GRADUATE STUDIES

I N M AT H E M AT I C S 244

Introduction
to Complex
Manifolds
John M. Lee
 10.1090/gsm/244

Introduction
to Complex
Manifolds
GRADUATE STUDIES
I N M AT H E M AT I C S 244

Introduction
to Complex
Manifolds

John M. Lee
 EDITORIAL COMMITTEE
Matthew Baker
Marco Gualtieri (Chair)
Sean T. Paul
Natasa Pavlovic
Rachel Ward

2020 Mathematics Subject Classification. Primary 32Qxx; Secondary 32-01, 32Q15,

53-01, 53C55, 53C56.

For additional information and updates on this book, visit

[Link]/bookpages/gsm-244

Library of Congress Cataloging-in-Publication Data

Names: Lee, John M., 1950- author.
Title: Introduction to complex manifolds / John M. Lee.
Description: Providence, Rhode Island : American Mathematical Society, [2024] | Series: Gradu-
ate studies in mathematics, 1065-7339 ; volume 244 | Includes bibliographical references and
index.
Identifiers: LCCN 2024011524 | ISBN 9781470476953 (hardcover) | ISBN 9781470477820 (paper-
back) | 9781470477813 (ebook)
Subjects: LCSH: Complex manifolds. | Manifolds (Mathematics) | AMS: Several complex vari-
ables and analytic spaces – Complex manifolds. | Several complex variables and analytic spaces
– Instructional exposition (textbooks, tutorial papers, etc.). | Several complex variables and
analytic spaces – Complex manifolds – Kähler manifolds. | Differential geometry – Instruc-
tional exposition (textbooks, tutorial papers, etc.). | Differential geometry – Global differential
geometry – Hermitian and Kählerian manifolds. | Differential geometry – Global differential
geometry – Other complex differential geometry.
Classification: LCC QA649 .L44 2024 | DDC 515/.946–dc23/eng/20240409
LC record available at [Link]

Copying and reprinting. Individual readers of this publication, and nonprofit libraries acting
for them, are permitted to make fair use of the material, such as to copy select pages for use
in teaching or research. Permission is granted to quote brief passages from this publication in
reviews, provided the customary acknowledgment of the source is given.
Republication, systematic copying, or multiple reproduction of any material in this publication
is permitted only under license from the American Mathematical Society. Requests for permission
to reuse portions of AMS publication content are handled by the Copyright Clearance Center. For
more information, please visit [Link]/publications/pubpermissions.
Send requests for translation rights and licensed reprints to reprint-permission@[Link].
c 2024 by the American Mathematical Society. All rights reserved.

The American Mathematical Society retains all rights
except those granted to the United States Government.
Printed in the United States of America.

∞ The paper used in this book is acid-free and falls within the guidelines
established to ensure permanence and durability.
Visit the AMS home page at [Link]
10 9 8 7 6 5 4 3 2 1 29 28 27 26 25 24
Contents

Preface ix

Chapter 1. The Basics 1

Definitions 1
Holomorphic Maps 7
Covering Manifolds and Quotient Manifolds 8
Some Complex Analysis 12
The Complexified Tangent and Cotangent Bundles 23
Almost Complex Structures 34
Problems 41

Chapter 2. Complex Submanifolds 45

Variations on the Inverse Function Theorem 45
Complex Submanifolds 48
Complex Submanifolds of Projective Spaces 52
The Holomorphic Embedding Problem 64
Problems 68

Chapter 3. Holomorphic Vector Bundles 71

Holomorphic Bundle Tools 71
Holomorphic Line Bundles 81
Line Bundles over Projective Space 84
Applications of Holomorphic Line Bundles 90
Problems 99

v
vi Contents

Chapter 4. The Dolbeault Complex 103

Decomposing Differential Forms by Type 103
A Poincaré Lemma for the Dolbeault Operator 109
Bundle-Valued Forms 114
Problems 118
Chapter 5. Sheaves 121
Definitions 121
The Étalé Space of a Presheaf 127
Exact Sequences of Sheaves 137
Problems 144
Chapter 6. Sheaf Cohomology 147
Definitions 147
The Long Exact Cohomology Sequence 154
Acyclic Resolutions 159
Sheaf Cohomology and Singular Cohomology 164
Applications of Sheaf Cohomology 173
Other Sheaf Cohomology Theories 183
Problems 189
Chapter 7. Connections 193
Connections on Complex Vector Bundles 193
Curvature 198
The First Real Chern Class 202
The Chern Connection 207
Problems 218
Chapter 8. Hermitian and Kähler Manifolds 221
Hermitian Metrics on the Tangent Bundle 221
Kähler Metrics 226
Examples of Kähler Metrics 234
Curvature of Kähler Metrics 239
Ricci and Scalar Curvatures 245
Problems 251
Contents vii

Chapter 9. Hodge Theory 255

The Hodge Inner Product 256
Elliptic Differential Operators 265
Hodge Theory on Riemannian Manifolds 269
Hodge Theory on Complex Manifolds 274
Hodge Theory on Kähler Manifolds 281
Applications of Hodge Theory 287
Problems 308
Chapter 10. The Kodaira Embedding Theorem 315
Preliminaries 315
The Kodaira Vanishing Theorem 317
Proof of the Embedding Theorem 330
Applications of the Embedding Theorem 335
Problems 337
Bibliography 343
Notation Index 347
Subject Index 351
Preface

Complex manifold theory is one of the most beautiful branches of geometry, in

which algebraic topology, differential geometry, algebraic geometry, homological
algebra, complex analysis, and partial differential equations (PDEs) come together
in deep and surprising ways to create a rich theory. This book is meant to be an
introduction to the concepts, techniques, and principal results about complex man-
ifolds (mainly compact ones), aimed primarily at graduate students and researchers
who already have a solid background in differential geometry.
What are complex manifolds? They are defined in exactly the same way as

ues in ℂ𝑛 and to overlap holomorphically. This might sound like a minor tweak to
smooth manifolds, except the local coordinate charts are required to take their val-

the definition of smooth manifolds, but in fact the requirement of holomorphicity

changes everything. For example, on a connected compact complex manifold, the
only global holomorphic functions are the constants, and the space of holomorphic
sections of a holomorphic vector bundle is always finite-dimensional. Whereas ev-

certain complex manifolds can be holomorphically embedded in ℂ𝑛 or in complex

ery smooth manifold can be smoothly embedded in some Euclidean space, only

projective space. There is a deep interplay between differential geometry and com-
plex analysis, especially for Kähler manifolds, the ones on which the metric struc-
ture and the holomorphic structure play together nicely.
Complex manifolds have profound applications in many areas of mathematics.
Here are a few examples:

• Riemann surfaces (1-dimensional complex manifolds) are essential for un-

derstanding global properties of holomorphic functions in one complex
variable.
• Complex surfaces (2-dimensional complex manifolds) play a central role
in attempts to classify 4-dimensional smooth manifolds.

ix
x Preface

• Complex manifolds defined by algebraic equations are among the central

objects of interest in algebraic geometry, and the study of their differential
geometry has contributed important advances in algebraic geometry.

• Calabi–Yau manifolds are complex manifolds that play a crucial role in

string theory.

Because complex manifold theory is so intimately connected with algebraic ge-

ometry and complex analysis as well as differential geometry, there are many paths
one might follow to get to know the subject. This book is about the differential
geometry of complex manifolds, so the techniques and results in it are all firmly
situated in differential geometry. Although I survey quite a few of the connections
with other subjects, especially complex algebraic geometry, my choice of topics
is heavily influenced by my desire to focus on techniques that will be familiar to
those with a good background in differential geometry but not necessarily in com-
mutative algebra or analysis of several complex variables. I have tried to include
enough points of contact with algebraic geometry that the book can serve as a useful
jumping-off point for differential geometers who decide to delve into the algebraic
geometry literature, while at the same time serving to introduce algebraic geometers
and complex analysts to a differential-geometric viewpoint on their subject.
I have chosen to use the Kodaira embedding theorem—which characterizes
those compact complex manifolds that admit holomorphic embeddings into pro-
jective spaces—as a unifying theme for the book, because it draws on most of the
important techniques in complex manifold theory and it illustrates one of the most
profound differences between smooth manifolds and complex ones. Many of the
definitions, theorems, and techniques introduced in the book are collected together
to lay the groundwork for proving that profound theorem. But not everything is
here for that purpose—I also hope to offer readers a strong general background that
will prepare them for more advanced study in any aspect of complex geometry.
There are many other good introductory books on complex manifolds. Some
excellent examples are [Bal06, Dem12, Huy05, Mor07, Wel08, Zhe00] and the first
few chapters of [GH94]. What distinguishes this book is an approach that readers
of my previous graduate texts will find familiar—instead of aiming for comprehen-
sive coverage of all the aspects of the subject (which would be impossible in any
case), I aim for two overarching goals: first, to make the explanations of definitions
and concepts user-friendly, well motivated, and accessible; and second, to write the
proofs with enough detail and rigor that students will hopefully not be left wonder-
ing how to bridge the gaps. This approach results in explanations that may be more
wordy than some readers are used to; but in my experience it really helps beginners
start to feel comfortable with a new subject.
Preface xi

Prerequisites
The main prerequisite is familiarity with the foundational results on topolog-
ical, smooth, and Riemannian manifolds. Because this subject draws on so many
of those results, there would be no point in trying to summarize all the requisite
differential-geometric background here. All of the background material on man-
ifolds that a reader needs to understand this book (and more) is contained in my
three previous graduate textbooks [LeeTM, LeeSM, LeeRM], and I draw freely on
them throughout this book (with specific references whenever appropriate).
Familiarity with elementary complex analysis is also a prerequisite, but only
at the level of a typical undergraduate course on complex analysis in one variable.
Any decent undergraduate complex analysis textbook will serve as a reference, such
as [BC13, MH98, Gam01].
Beyond these subjects, the reader should also have a basic familiarity with al-
gebraic topology—particularly singular homology and cohomology at the level of
[Hat02] or [Mun84]. I give references for the main results that I use in the text.
For readers who are well versed in the prerequisite material, this book should
be essentially self-contained, with one major exception: all of the results on Hodge
theory rest on a fundamental Fredholm theorem for elliptic partial differential equa-
tions (Thm. 9.14), which is stated here without proof because developing the ma-
chinery for proving it would carry us too far afield into the weeds of PDE theory.
The theorem is easy to state and easy to use, so readers can accept it on faith; or,
for those who are curious about the proof, I offer several references where proofs
can be found.

Exercises and Problems

Like my other graduate texts, this book includes both exercises (integrated into
the text) and problems (collected at the ends of the chapters). The exercises are
mostly routine verifications, and are there to provide the reader with opportunities
to check how well they have digested the material; while the problems are mostly
more difficult (some considerably so), and are designed to challenge the reader to
grapple more deeply with the ideas in the text. As was the case with my previous
books, I have not and do not intend to prepare written solutions to the problems
or the exercises, because I do not want to deprive readers of the opportunity to get
“stuck” on a problem and do the productive work of finding their own ways forward.
In any case, most of these are not problems that have a single “right answer.”

Typographical Conventions
This book generally follows the same typographical conventions as my previ-
ous graduate texts. Mathematical terms are typeset in bold italics when they are
officially defined, to make them easy to spot on the page. The exercises in the text
xii Preface

are indicated with the symbol ►, and numbered consecutively with the theorems to
make them easy to find. The symbol □ marks the ends of proofs, and also marks the
ends of statements of corollaries that follow so easily that they do not need proofs.
The symbol // marks the ends of numbered examples. End-of-chapter problems
are numbered 1-1, 1-2, 1-3, etc., with hyphens instead of dots, to make it easier to
distinguish problem references from exercise references.

Acknowledgements
As always, I owe a lot to my students who have given me feedback on early
drafts of this text, especially Shahriar Talebi. I also want to thank Jim Isenberg,
who gave the whole book a close reading and contributed invaluable suggestions.
Finally, I want to thank Ina Mette at the AMS, who has been wonderfully encour-
aging and patient as I have struggled to bring this project to fruition.
I welcome feedback from readers about any aspects of the book, especially if
you find mistakes or unclear passages. There will be an updated list of corrections
on my website. I hope you enjoy the book.

Seattle, Washington, USA John M. Lee

johnmlee@[Link]
[Link]/~lee/
 10.1090/gsm/244/01

Chapter 1

The Basics

If you are familiar with the prerequisites for reading this book, you are proba-
bly already familiar with the notion of a smooth manifold—a topological mani-
fold equipped with an atlas of coordinate charts whose transition functions are all
smooth. (Precise definitions will be found farther down in this chapter.)
You may also have encountered variations on that theme—different classes

charts. For example, a 𝑪 𝒌 manifold is one equipped with an atlas whose transi-
of manifolds that can be defined by modifying the compatibility condition for

tion functions are all of class 𝐶 𝑘 (meaning 𝑘 times continuously differentiable),

and a real-analytic manifold is one with an atlas whose transition functions are all
real-analytic (meaning they are equal to the sum of a convergent power series in a
neighborhood of each point).
Another variation on that theme, and the one to which this book is devoted, is
a complex manifold—this is a topological manifold equipped with an atlas whose
transition functions are all holomorphic. While the other classes of manifolds men-
tioned above are really just slight variations on the theme of smooth manifolds, it
turns out that nearly everything changes when we move into the holomorphic cate-
gory, as you will soon see. That is why the subject of complex manifolds is worth
an entire book of its own.
In this chapter we introduce the main definitions, and describe some examples
and basic properties of complex manifolds.

Definitions
The most basic type of manifold is a topological manifold: this is a second-

neighborhood homeomorphic to an open subset of ℝ𝑛 for some fixed 𝑛, called the

countable Hausdorff topological space with the property that every point has a

1
2 1. The Basics

dimension of the manifold. (In this book, all manifolds are understood to be mani-
folds without boundary unless otherwise specified.)
By adding extra structure to a topological manifold, we can obtain other types
of manifolds. Differential geometry is concerned primarily with smooth mani-

as follows: If 𝑀 is a topological manifold of dimension 𝑛, a coordinate chart

folds, which are topological manifolds endowed with smooth structures, defined

(often called just a chart) for 𝑀 is a pair (𝑈 , 𝜑), where 𝑈 is an open subset of
𝑀 and 𝜑 is a homeomorphism from 𝑈 to an open subset of ℝ𝑛 . An atlas for
𝑀 is a collection of charts whose domains cover 𝑀. Given two charts (𝑈 , 𝜑) and
(𝑉 , 𝜓) with overlapping domains, their transition functions are the composite maps
𝜓 ∘𝜑−1 ∶ 𝜑(𝑈 ∩𝑉 ) → 𝜓(𝑈 ∩𝑉 ) and their inverses 𝜑∘𝜓 −1 ∶ 𝜓(𝑈 ∩𝑉 ) → 𝜑(𝑈 ∩𝑉 ).

transition functions are smooth as maps between open subsets of ℝ𝑛 . (Here and
Two charts are said to be smoothly compatible if their domains are disjoint or their

throughout the book, smooth means infinitely differentiable or of class 𝐶 ∞ .) A

smooth atlas for 𝑀 is an atlas with the property that any two charts in the atlas
are smoothly compatible with each other. Finally, a smooth structure for 𝑀 is a

smooth atlas; to say that 𝒜 is a maximal smooth atlas just means that every chart
smooth atlas that is maximal, meaning that it is not properly contained in any larger

that is smoothly compatible with every chart in 𝒜 is already in 𝒜 .

The definition of a complex manifold is, at first glance, just a minor modifica-
tion of the definition of smooth manifolds. The main change is that we require each
transition function to be holomorphic, meaning that it is continuous and each of its

to each of the independent complex variables 𝑧1 , … , 𝑧𝑛 . (We will explore proper-

complex-valued component functions has complex partial derivatives with respect

ties of holomorphic functions in more depth below; for now, it suffices to know that
they are smooth and that compositions of holomorphic functions are holomorphic.)

choose the following standard identification between ℝ2𝑛 and ℂ𝑛 :

To apply this requirement to the transition functions for topological manifolds, we

(𝑥1 , 𝑦1 , … , 𝑥𝑛 , 𝑦𝑛 ) ↔ (𝑥1 + 𝑖𝑦1 , … , 𝑥𝑛 + 𝑖𝑦𝑛 ).

(As in [LeeSM] and [LeeRM], we index coordinate functions with upper indices
to be consistent with the Einstein summation convention, described later in this

open subsets of ℝ2𝑛 is holomorphic.

chapter.) With this identification, it makes sense to ask whether a map between

Now suppose 𝑀 is a 2𝑛-dimensional topological manifold. If (𝑈 , 𝜑) and (𝑉 , 𝜓)

are two coordinate charts for 𝑀, we say they are holomorphically compatible if
𝑈 ∩ 𝑉 = ∅ or both transition functions are holomorphic under our standard iden-
tification of 𝜑(𝑈 ∩ 𝑉 ) and 𝜓(𝑈 ∩ 𝑉 ) as open subsets of ℂ𝑛 . A holomorphic atlas
for 𝑀 is an atlas with the property that any two charts in the atlas are holomorphi-
cally compatible with each other, and a holomorphic structure for 𝑀 is a maximal
holomorphic atlas. An 𝒏-dimensional complex manifold (or holomorphic mani-
fold) is a topological manifold of dimension 2𝑛 endowed with a given holomorphic
Definitions 3

structure. A complex manifold of dimension 1 is called a complex curve, and one

of dimension 2 is called a complex surface. A complex manifold of dimension 3

sary to distinguish between the dimension of an 𝑛-dimensional complex manifold

or higher is sometimes called a complex threefold, fourfold, etc. When it is neces-

and the dimension of its underlying topological 2𝑛-manifold, we call 𝑛 the complex
dimension (denoted by dimℂ 𝑀) and 2𝑛 the real dimension (denoted by dimℝ 𝑀).

coordinate chart, and the complex-valued coordinate functions (𝑧1 , … , 𝑧𝑛 ) (where

Any one of the charts in the maximal holomorphic atlas is called a holomorphic

𝑧𝑗 = 𝑥𝑗 + 𝑖𝑦𝑗 ) are called holomorphic coordinates. We denote the complex conju-

gate of 𝑧𝑗 by 𝑧𝑗 = 𝑥𝑗 − 𝑖𝑦𝑗 .
Holomorphic structures on manifolds are traditionally called complex struc-
tures, but that term risks confusion with complex structures on vector bundles, to
be discussed below.
Because all holomorphic functions are smooth (see Thm. 1.21 below), a holo-

on 𝑀; thus every complex manifold is also a smooth manifold in a canonical way.

morphic atlas is also a smooth atlas and thus determines a unique smooth structure

On the other hand, it is important to note that a given even-dimensional smooth

manifold may have many different holomorphic structures that induce the given
smooth structure (see Problem 1-4), or it may have none at all. The simplest exam-

is 𝕊4 ; see the discussion following Theorem 1.63 for more detail.

ple of an even-dimensional smooth manifold that carries no holomorphic structure

Proposition 1.1. Let 𝑀 be a topological manifold.

(a) Every holomorphic atlas 𝒜 for 𝑀 is contained in a unique maximal holo-
morphic atlas, called the holomorphic structure determined by 𝓐.
(b) Two holomorphic atlases for 𝑀 determine the same holomorphic struc-
ture if and only if their union is a holomorphic atlas.

Proof. The proof is essentially identical to that of its smooth counterpart [LeeSM,
Prop. 1.17]. □

To turn a set into a complex manifold using the definitions directly, it would
be necessary to go through the separate steps of constructing a topology, verifying
that it is a manifold, and then constructing a holomorphic structure for it. But in
most cases the following shortcut can be used.
Lemma 1.2 (Complex Manifold Chart Lemma). Let 𝑀 be a set, and suppose we
are given a collection {𝑈𝛼 }𝛼∈𝐴 of subsets of 𝑀 together with maps 𝜑𝛼 ∶ 𝑈𝛼 → ℂ𝑛 ,
such that the following properties are satisfied:
(i) For each 𝛼, 𝜑𝛼 is a bijection between 𝑈𝛼 and an open subset 𝜑𝛼 (𝑈𝛼 ) ⊆ ℂ𝑛 .
(ii) For each 𝛼 and 𝛽, the sets 𝜑𝛼 (𝑈𝛼 ∩ 𝑈𝛽 ) and 𝜑𝛽 (𝑈𝛼 ∩ 𝑈𝛽 ) are open in ℂ𝑛 .
4 1. The Basics

(iii) When 𝑈𝛼 ∩ 𝑈𝛽 ≠ ∅, the map 𝜑𝛽 ∘ 𝜑−1

𝛼 ∶ 𝜑𝛼 (𝑈𝛼 ∩ 𝑈𝛽 ) → 𝜑𝛽 (𝑈𝛼 ∩ 𝑈𝛽 ) is
holomorphic.
(iv) Countably many of the sets 𝑈𝛼 cover 𝑀.
(v) Whenever 𝑝, 𝑞 are distinct points in 𝑀, either there exists some 𝑈𝛼 con-
taining both 𝑝 and 𝑞 or there exist disjoint sets 𝑈𝛼 , 𝑈𝛽 with 𝑝 ∈ 𝑈𝛼 and
𝑞 ∈ 𝑈𝛽 .

Then 𝑀 has a unique structure as a complex manifold such that each (𝑈𝛼 , 𝜑𝛼 ) is a
holomorphic chart.

► Exercise 1.3. Prove this lemma by verifying that the proof of Lemma 1.35 of
[LeeSM] goes through in this setting.

Some Examples
Before we go much further, we should have a few examples of complex mani-
folds to think about. We will introduce many more examples in Chapter 2.

Example 1.4 (Complex 𝒏-Space). It follows from Proposition 1.1(a) that ℂ𝑛 has a

the single coordinate chart (ℂ𝑛 , Idℂ𝑛 ). Similarly, the canonical holomorphic struc-
canonical holomorphic structure determined by the holomorphic atlas consisting of

ture on every open subset 𝑈 ⊆ ℂ𝑛 is defined by the single chart (𝑈 , Id𝑈 ). When
working with ℂ, ℂ𝑛 , or their open subsets, we always use this holomorphic struc-
ture, typically without further comment. Here are some specific open subsets that
will play important roles in what follows:

• For any 𝑝 ∈ ℂ𝑛 and any 𝑟 > 0, the (open) ball of radius 𝒓 around 𝒑 is
the set 𝐵𝑟 (𝑝) = {𝑧 ∈ ℂ𝑛 ∶ |𝑧 − 𝑝| < 𝑟}, where | ⋅ | denotes the norm
associated with the Euclidean inner product on ℂ𝑛 ≈ ℝ2𝑛 , which can
be written in complex coordinates as ⟨𝑧, 𝑤⟩ = 𝑧 ⋅ 𝑤 = ∑𝑛𝑗=1 𝑧𝑗 𝑤𝑗 . The
unit ball of real dimension 2𝑛, denoted by 𝔹2𝑛 , is the open ball of radius
1 about the origin in ℂ𝑛 .
• An open ball in ℂ is called a disk, and the notation is modified accordingly.
Thus 𝐷𝑟 (𝑝) represents the disk of radius 𝑟 about 𝑝 ∈ ℂ, and the unit disk
is the disk 𝐷1 (0), denoted by 𝔻.
• A polydisk is a Cartesian product of open disks, that is, an open subset of
the form 𝐷𝑟1 (𝑝1 ) × ⋯ × 𝐷𝑟𝑛 (𝑝𝑛 ) ⊆ ℂ𝑛 for a point 𝑝 = (𝑝1 , … , 𝑝𝑛 ) ∈ ℂ𝑛
and positive real numbers 𝑟1 , … , 𝑟𝑛 . When the radii are all equal, we use
the notation 𝐷𝑟𝑛 (𝑝) for the polydisk 𝐷𝑟 (𝑝1 ) × ⋯ × 𝐷𝑟 (𝑝𝑛 ). //

Example 1.5 (Open Submanifolds). Somewhat more generally, if 𝑀 is a complex

𝑛-manifold and 𝑈 is an open subset of 𝑀, we can define a canonical holomorphic
Definitions 5

structure on 𝑈 consisting of all holomorphic charts for 𝑀 whose domains are con-
tained in 𝑈 . With this holomorphic structure, 𝑈 is a complex 𝑛-manifold, called an
open submanifold of 𝑴. //
Example 1.6 (Complex Vector Spaces). If 𝑉 is a finite-dimensional complex
vector space, any choice of ordered basis (𝑏1 , … , 𝑏𝑛 ) defines an isomorphism
𝐵 ∶ ℂ𝑛 → 𝑉 by
(1.1) 𝐵(𝑧1 , … , 𝑧𝑛 ) = 𝑧𝑗 𝑏𝑗 .
(Here and throughout the book, we use the Einstein summation convention: each
index name that appears twice in the same monomial term, once as an upper index

index, typically from 1 to the dimension of the space. In formula (1.1), since 𝑉 has
and once as a lower one, is understood to be summed over all possible values of that

dimension 𝑛, the implied summation is from 1 to 𝑛.) Interpreting 𝐵 −1 as a global

chart thus defines a holomorphic structure on 𝑉 . Since the transition map between
any two such charts is an invertible complex-linear transformation and therefore

basis. We will call this the standard holomorphic structure on 𝑽 .

holomorphic along with its inverse, this structure is independent of the choice of
//
Example 1.7 (0-Manifolds). A topological 0-manifold is just a countable discrete
space. Each point has a unique map to ℂ0 = {0}, and the transition functions be-
tween these maps are vacuously holomorphic, so every 0-manifold has a canonical
holomorphic structure. //
Example 1.8 (Product Manifolds). If 𝑀1 , … , 𝑀𝑘 are complex manifolds, their
Cartesian product 𝑀1 × ⋯ × 𝑀𝑘 (with the product topology) is a complex manifold
whose dimension is the sum of the dimensions of the factors, with products of
holomorphic coordinate maps providing holomorphic coordinates. //
Example 1.9 (Complex Projective Spaces). The next examples are, after ℂ𝑛 itself,
the most important complex manifolds of all. For any nonnegative integer 𝑛, we
define the complex projective space of dimension 𝒏, denoted by ℂℙ𝑛 , to be the
set of complex 1-dimensional subspaces of ℂ𝑛+1 , which we can identify with the
quotient of ℂ𝑛+1 ∖ {0} by the equivalence relation defined by 𝑤 ∼ 𝑤′ if and only if
𝑤′ = 𝜆𝑤 for some nonzero complex number 𝜆. We endow ℂℙ𝑛 with the quotient
topology. By this definition, ℂℙ0 is a single point.
We denote the equivalence class of a point 𝑤 = (𝑤0 , 𝑤1 , … , 𝑤𝑛 ) ∈ ℂ𝑛+1 ∖ {0}
by [𝑤] = [𝑤0 , … , 𝑤𝑛 ]. The complex numbers (𝑤0 , … , 𝑤𝑛 ) are traditionally called
homogeneous coordinates of the point [𝑤]; but be careful about using this termi-

[𝑤] is represented by any homogeneous coordinates of the form (𝜆𝑤0 , … , 𝜆𝑤𝑛 )

nology, because they are not actually coordinates in the usual sense. The same point

with 𝜆 ≠ 0, so there is not a one-to-one correspondence between points and homo-

We can construct honest coordinates for ℂℙ𝑛 as follows. For each 𝛼 = 0, … , 𝑛,

geneous coordinates, even in a small neighborhood of a point.

let 𝑈𝛼 ⊆ ℂℙ𝑛 be the open subset 𝑈𝛼 = {[𝑤] ∈ ℂℙ𝑛 ∶ 𝑤𝛼 ≠ 0}, and define a map
6 1. The Basics

𝜑𝛼 ∶ 𝑈𝛼 → ℂ𝑛 by
𝑤0 𝑤𝛼−1 𝑤𝛼+1 𝑤𝑛
𝜑𝛼 ([𝑤0 , … , 𝑤𝑛 ]) = , … , , , … , .
( 𝑤𝛼 𝑤𝛼 𝑤𝛼 𝑤𝛼 )
It is continuous by the characteristic property of the quotient topology [LeeTM,
Thm. 3.70], and it is a homeomorphism because it has a continuous inverse given

𝜑−1
𝛼 (𝑧 , … , 𝑧 ) = [𝑧 , … , 𝑧 , 1, 𝑧𝛼 , … , 𝑧𝑛 ].
by
1 𝑛 1 𝛼−1

Thus each (𝑈𝛼 , 𝜑𝛼 ) is a coordinate chart, called affine coordinates for ℂℙ𝒏 . Cou-
pled with the facts that ℂℙ𝑛 is Hausdorff and second-countable (Exercise 1.10),
this shows that ℂℙ𝑛 is a topological manifold of real dimension 2𝑛. It is com-

𝑞 ∶ 𝕊2𝑛+1 → ℂℙ𝑛 given by 𝑞(𝑤0 , … , 𝑤𝑛 ) = [𝑤0 , … , 𝑤𝑛 ], where 𝕊2𝑛+1 is the

pact and connected, because it is the image of the surjective continuous map

set of unit vectors in ℂ𝑛+1 .

For 𝛼 < 𝛽, the transition function between these charts can be computed ex-
plicitly as
𝑧1 ̂
𝑧𝛼 1 𝑧𝑛
𝜑𝛼 ∘ 𝜑−1
𝛽 (𝑧 , … , 𝑧 ) =
1 𝑛
, … , , … , , … , ,
( 𝑧𝛼 𝑧𝛼 𝑧𝛼 𝑧𝛼 )
where the hat indicates that the term in position 𝛼 is omitted, and the 1/𝑧𝛼 term is
in position 𝛽; the formula for 𝛼 > 𝛽 is similar. These transition functions are all
holomorphic, so they turn ℂℙ𝑛 into a complex manifold of dimension 𝑛. //

► Exercise 1.10. Verify that ℂℙ𝑛 is Hausdorff and second-countable.

Example 1.11 (Projectivization of a Vector Space). For some purposes, it is use-

place of ℂ𝑛+1 itself. Suppose 𝑉 is an 𝑛-dimensional complex vector space with

ful to construct projective spaces starting with different complex vector spaces in

𝑛 > 0. The projectivization of 𝑽 , denoted by ℙ(𝑉 ), is the set of 1-dimensional

complex subspaces of 𝑉 , endowed with the quotient topology obtained from the
equivalence relation on 𝑉 ∖ {0} given by 𝑣1 ∼ 𝑣2 if 𝑣2 = 𝜆𝑣1 for some 𝜆 ∈ ℂ ∖ {0}.
A choice of basis for 𝑉 yields an isomorphism 𝑉 ≅ ℂ𝑛 that descends to a bijection
ℙ(𝑉 ) → ℂℙ𝑛−1 , which we can use to give ℙ(𝑉 ) the structure of a complex mani-
fold. In the next chapter, we will see that the holomorphic structure obtained in this
way is independent of the choice of basis (see Exercise 2.10). //
Example 1.12 (Complex Grassmannians). Suppose 𝑉 is an 𝑛-dimensional com-
plex vector space with 𝑛 > 0, and 𝑘 is a nonnegative integer less than or equal to
𝑛. Let G𝑘 (𝑉 ) be the set of 𝑘-dimensional complex-linear subspaces of 𝑉 , called a
complex Grassmannian. (The case 𝑘 = 1 is exactly the projective space ℙ(𝑉 ).)
We can construct complex coordinates on G𝑘 (𝑉 ) as follows. Choose a subspace
𝑃 ⊆ 𝑉 of dimension 𝑘 and a complementary (𝑛 − 𝑘)-dimensional subspace 𝑄, and
write 𝑉 = 𝑃 ⊕ 𝑄. Then the graph of each complex-linear map 𝑋 ∶ 𝑃 → 𝑄 is a
Holomorphic Maps 7

𝑘-dimensional subspace Γ(𝑋) ⊆ 𝑉 , and every subspace whose intersection with

𝑄 is trivial is the graph of a unique such map. Let 𝑈𝑄 ⊆ G𝑘 (𝑉 ) denote the set of
such subspaces. By choosing bases for 𝑃 and 𝑄, we obtain a bijection from 𝑈𝑄 to
the vector space M((𝑛 − 𝑘) × 𝑘, ℂ) of complex (𝑛 − 𝑘) × 𝑘 matrices, whose matrix
entries we can use as coordinates on 𝑈𝑄 . The argument in [LeeSM, Example 1.36]
(adapted in an obvious way to the complex case) shows that when two such charts
overlap, the matrix entries in the new chart are rational functions of the original
ones, so any two such charts overlap holomorphically. The arguments of that ex-

G𝑘 (𝑉 ) is a complex manifold of dimension (𝑛 − 𝑘)𝑘. Problem 1-5 shows that it is

ample also show that hypotheses (iv) and (v) of the chart lemma are satisfied, so

compact. //

Holomorphic Maps

defines smooth maps between smooth manifolds: if 𝑀 and 𝑁 are complex mani-
We define holomorphic maps between complex manifolds in the same way as one

folds, a holomorphic map from 𝑀 to 𝑁 is a map 𝑓 ∶ 𝑀 → 𝑁 with the property

that for every 𝑝 ∈ 𝑀 there exist holomorphic coordinate charts (𝑈 , 𝜑) for 𝑀 and
(𝑉 , 𝜓) for 𝑁 whose domains contain 𝑝 and 𝑓 (𝑝), respectively, such that 𝑓 (𝑈 ) ⊆ 𝑉
and the composite map 𝜓 ∘ 𝑓 ∘ 𝜑−1 is holomorphic as a map from 𝜑(𝑈 ) to 𝜓(𝑉 ).
The function 𝑓̂ = 𝜓 ∘ 𝑓 ∘ 𝜑−1 is called the coordinate representation of 𝒇 with
respect to the given holomorphic coordinates. As is the case in smooth manifold

identify an open subset of a manifold with an open subset of ℂ𝑛 , and uses the same
theory (see [LeeSM, pp. 15–16]), one often uses a coordinate map to temporarily

When the codomain of a map 𝑓 is ℂ𝑘 (or an open subset of ℂ𝑘 ) with its canon-
notation for a map and its coordinate representation.

coordinate chart on ℂ𝑘 , so being holomorphic is equivalent to the requirement that

ical holomorphic structure, we can always use the identity map as a holomorphic

for each 𝑝 ∈ 𝑀, there is a holomorphic chart (𝑈 , 𝜑) for 𝑀 whose domain contains

𝑝 such that 𝑓 ∘𝜑−1 is holomorphic from 𝜑(𝑈 ) to ℂ𝑘 . It is standard practice to reserve

subsets of ℂ (scalar-valued holomorphic functions) or ℂ𝑘 (vector valued holomor-

the term holomorphic function for holomorphic maps whose codomains are open

phic functions); the terms holomorphic map and holomorphic mapping can refer to

If 𝑀 is a complex manifold, the notation 𝒪(𝑀) means the set of all holomor-
maps between arbitrary complex manifolds.

phic functions from 𝑀 to ℂ. This applies, in particular, to any open submanifold

of 𝑀: if 𝑈 ⊆ 𝑀 is open, 𝒪(𝑈 ) is the set of holomorphic functions from 𝑈 to ℂ.
A bijective holomorphic map with holomorphic inverse is called a biholomor-

morphism. More generally, a map 𝐹 ∶ 𝑀 → 𝑁 is called a local biholomorphism

phism, and a biholomorphism from a complex manifold to itself is called an auto-

if every 𝑝 ∈ 𝑀 has a neighborhood 𝑈 such that 𝐹 |𝑈 is a biholomorphism onto an

open subset of 𝑁.
8 1. The Basics

The following facts about holomorphic maps are proved just like their smooth
analogues [LeeSM, Props. 2.6 and 2.10 and Example 2.14(b)].
Proposition 1.13.
(a) The restriction of a holomorphic map to an open subset is holomorphic.
(b) If a map 𝑓 has the property that each point in the domain has a neighbor-
hood 𝑈 on which the restriction 𝑓 |𝑈 is holomorphic, then 𝑓 is holomor-
phic.
(c) Every constant map between complex manifolds is holomorphic.
(d) The identity map of every complex manifold is holomorphic.
(e) The inclusion map of every open submanifold is holomorphic.
(f) Every holomorphic coordinate chart is a biholomorphism onto its image.
(g) Every composition of holomorphic maps between complex manifolds is
holomorphic.

phism between them. For example, if 𝑉 is an 𝑛-dimensional complex vector space,

Two complex manifolds are said to be biholomorphic if there is a biholomor-

any choice of basis determines a complex-linear isomorphism between 𝑉 and ℂ𝑛 ,

so all such vector spaces are biholomorphic to ℂ𝑛 . Similarly, a choice of basis yields
a biholomorphism between ℙ(𝑉 ) and ℂℙ𝑛−1 , and between G𝑘 (𝑉 ) and G𝑘 (ℂ𝑛 ) for
each 𝑘. It is easy to check that being biholomorphic is an equivalence relation on the
class of all complex manifolds. The main subject matter of this book is properties
of complex manifolds that are preserved by biholomorphisms.
Because holomorphic maps are smooth, biholomorphic manifolds are automat-
ically diffeomorphic. However, the converse might not be true: Example 1.31 and
Problem 1-4 describe complex manifolds that are diffeomorphic but not biholomor-
phic.

Covering Manifolds and Quotient Manifolds

ones. Recall that a covering map is a surjective continuous map 𝜋 ∶ 𝑀 → 𝑁

In this section, we discuss some ways to produce new complex manifolds from old

point of 𝑁 has a neighborhood 𝑈 that is evenly covered, meaning that 𝜋 −1 (𝑈 ) is a

between connected and locally path-connected topological spaces such that every

cally onto 𝑈 by 𝜋. A covering map 𝜋 ∶ 𝑀 → 𝑁 is said to be normal if for some

disjoint union of connected open subsets each of which is mapped homeomorphi-

𝑥 ∈ 𝑀, the induced subgroup 𝜋∗ (𝜋1 (𝑀, 𝑥)) ⊆ 𝜋1 (𝑁, 𝜋(𝑥)) is a normal subgroup
(meaning it is invariant under conjugation). Equivalently, 𝜋 is normal if the group
of covering automorphisms (homeomorphisms 𝜑 ∶ 𝑀 → 𝑀 satisfying 𝜋 ∘ 𝜑 = 𝜋)
acts transitively on each fiber 𝜋 −1 (𝑦). A discussion of the properties of covering
maps can be found in [LeeTM, Chaps. 11 & 12].
Covering Manifolds and Quotient Manifolds 9

Suppose 𝜋 ∶ 𝑀 → 𝑁 is a covering map. If 𝑀 and 𝑁 are smooth manifolds and

𝜋 is a local diffeomorphism, then it is called a smooth covering map. Properties of
smooth covering maps are discussed in [LeeSM, pp. 91–95]. Similarly, if 𝑀 and 𝑁
are complex manifolds and 𝜋 is a local biholomorphism, it is called a holomorphic
covering map.

► Exercise 1.14. Suppose 𝜋 ∶ 𝑀 → 𝑁 is a holomorphic covering map. Show

that every point of 𝑀 is in the image of a holomorphic local section of 𝜋, that
is, a holomorphic map 𝜎 ∶ 𝑈 → 𝑀 defined on an open subset 𝑈 ⊆ 𝑁 such that
𝜋 ∘ 𝜎 = Id𝑈 .

The next proposition shows that every covering space of a connected complex
manifold is a complex manifold in a natural way.

Suppose 𝑀 is a connected complex manifold and 𝜋 ∶ 𝐸 → 𝑀 is a (topological)

Proposition 1.15 (Coverings of Complex Manifolds are Complex Manifolds).

covering map. Then 𝐸 is a topological manifold and has a unique holomorphic

structure such that 𝜋 is a holomorphic covering map.

Proof. Proposition 4.40 in [LeeSM] shows that 𝐸 is a topological manifold and

has a unique smooth structure such that 𝜋 is a smooth covering map. We can de-
fine holomorphic charts on 𝐸 as follows: Given a point 𝑝 ∈ 𝐸, let 𝑈 be an evenly
covered neighborhood of 𝜋(𝑝). After shrinking 𝑈 if necessary, we can find a holo-
morphic coordinate map 𝜑 ∶ 𝑈 → ℂ𝑛 . Let 𝑈 ̃ be the connected component of
𝜋 (𝑈 ) containing 𝑝, and define 𝜑
−1
̃ = 𝜑 ∘ 𝜋∶ 𝑈 ̃ → ℂ𝑛 . The argument in the proof
of [LeeSM, Prop. 4.40] shows that when two such charts (𝑈 ̃, 𝜑
̃ ) and (𝑉̃ , 𝜓
̃ ) over-

𝜓
̃ −1 ∘ 𝜑̃ −1 = 𝜓 −1 ∘ 𝜑−1 , which in this case is holomorphic. Then 𝜋 is a local bi-
lap, in a neighborhood of each point the transition function can be expressed as

the holomorphic coordinates (𝑈 ̃, 𝜑

̃ ) on 𝐸 and (𝑈 , 𝜑) on 𝑀.
holomorphism because its coordinate representation is the identity with respect to

̃ is the same topological space 𝐸 with another holomorphic structure such

If 𝐸
that 𝜋 ∶ 𝐸 ̃ → 𝑀 is a holomorphic covering map, then because 𝜋 is a local biholo-
morphism, each of the charts constructed above must be a holomorphic chart for 𝐸, ̃
so the holomorphic structure of 𝐸 ̃ is the same as the one constructed above. □

ifolds covered by complex manifolds. Suppose Γ is a discrete Lie group (i.e., a

Under certain circumstances, we can also put holomorphic structures on man-

countable group with the discrete topology). Recall that an action of Γ on a man-
ifold 𝑀 is free if 𝑔 ⋅ 𝑥 = 𝑥 for some 𝑔 ∈ Γ and 𝑥 ∈ 𝑀 implies 𝑔 is the identity;
and it is proper if the map Γ × 𝑀 → 𝑀 × 𝑀 given by (𝑔, 𝑥) ↦ (𝑔 ⋅ 𝑥, 𝑥) is

[LeeSM, pp. 543–544].) If 𝑀 is a complex manifold, the action is holomorphic if

a proper map, meaning that the preimage of every compact set is compact. (See

the map 𝑥 ↦ 𝑔 ⋅ 𝑥 is holomorphic for each 𝑔 ∈ Γ.

10 1. The Basics

Theorem 1.16 (Holomorphic Quotient Manifold Theorem). Suppose Γ is a dis-

𝑀. Then the quotient space 𝑀/Γ has a unique complex manifold structure such
crete Lie group acting holomorphically, freely, and properly on a complex manifold

that the quotient map 𝑞 ∶ 𝑀 → 𝑀/Γ is a holomorphic normal covering map.

Proof. Smooth manifold theory shows that 𝑀/Γ has a unique smooth manifold
structure such that 𝑞 is a smooth normal covering map [LeeSM, Thm. 21.13]. To
define a complex manifold structure on 𝑀/Γ, let 𝑈 ⊆ 𝑀/Γ be any evenly covered
open set, and choose a smooth local section 𝜎 ∶ 𝑈 → 𝑀. Because 𝑀 is a complex
manifold, 𝜎(𝑈 ) has a covering by holomorphic charts (𝑈𝛼 , 𝜑𝛼 ), and for each such
chart we can define (𝜎 −1 (𝑈𝛼 ), 𝜑𝛼 ∘𝜎)) as a chart for 𝑀/Γ. For a fixed local section 𝜎,
all of these charts are holomorphically compatible with each other. If 𝜎 ̃∶ 𝑈 → 𝑀
is any other local section, there is an element 𝑔 ∈ Γ such that 𝜎 ̃(𝑥) = 𝑔 ⋅ 𝜎(𝑥) for
all 𝑥 ∈ 𝑈 ; and the fact that 𝑥 ↦ 𝑔 ⋅ 𝑥 is a biholomorphism of 𝑀 with inverse
𝑥 ↦ 𝑔 −1 ⋅ 𝑥 guarantees that the charts obtained from 𝜎 ̃ will be holomorphically
compatible with those obtained from 𝜎. □

A complex Lie group is a complex manifold 𝐺 endowed with a group structure

such that the multiplication map 𝑚 ∶ 𝐺 × 𝐺 → 𝐺 and the inversion map 𝑖 ∶ 𝐺 → 𝐺
are holomorphic. Here are some simple examples; we will see more in the next
chapter (see Example 2.26).
• Every countable discrete group is a 0-dimensional complex Lie group.
• Every finite-dimensional complex vector space is a complex Lie group
under addition.
• The group GL(𝑛, ℂ) of invertible 𝑛 × 𝑛 complex matrices is a complex
Lie group of dimension 𝑛2 , with the matrix entries as global holomorphic
coordinates. The component functions of the multiplication map are holo-
morphic polynomials in the matrix entries, and those of the inversion map
are holomorphic rational functions.
• Given any 𝑛-dimensional complex vector space 𝑉 , the group GL(𝑉 ) of
complex linear automorphisms of 𝑉 becomes a Lie group isomorphic to
GL(𝑛, ℂ) once we choose a basis for 𝑉 , and the resulting holomorphic
structure is independent of the choice of basis.
Corollary 1.17. Suppose 𝐺 is a connected complex Lie group and Γ ⊆ 𝐺 is a
discrete subgroup. The left coset space 𝐺/Γ is a complex manifold, and the quotient
map 𝜋 ∶ 𝐺 → 𝐺/Γ is a holomorphic normal covering map. If Γ is also a normal
subgroup, then 𝐺/Γ is a complex Lie group and 𝜋 is a group homomorphism.

Proof. The left coset space 𝐺/Γ is the quotient of 𝐺 by the action of Γ by right
translation. This action is holomorphic by the definition of a complex Lie group,
and the proof of Theorem 21.17 in [LeeSM] shows that it is free and proper. Thus
Covering Manifolds and Quotient Manifolds 11

Theorem 1.16 above shows that 𝐺/Γ has the structure of a complex manifold and 𝜋

If Γ is a normal subgroup, then elementary group theory shows that 𝐺/Γ is

is a holomorphic normal covering map.

a group and 𝜋 is a homomorphism. To see that the group operations in 𝐺/Γ are
holomorphic, just note that given any pair of points 𝑝, 𝑞 ∈ 𝐺/Γ, we can choose
neighborhoods 𝑈 of 𝑝 and 𝑉 of 𝑞 on which there exist holomorphic local sections
𝜎 ∶ 𝑈 → 𝐺 and 𝜏 ∶ 𝑉 → 𝐺. Then the multiplication map 𝑚 ̃ ∶ 𝐺/Γ × 𝐺/Γ → 𝐺/Γ
can be written in a neighborhood of (𝑝, 𝑞) as 𝜋 ∘ 𝑚 ∘ (𝜎 × 𝜏):
𝑚
𝐺 ×O 𝐺 /𝐺

𝜎×𝜏 𝜋
𝐺/Γ × 𝐺/Γ / 𝐺/Γ.
𝑚
̃


This is a composition of holomorphic maps and thus holomorphic. A similar argu-

ment applies to inversion. □

Example 1.18 (Complex Tori). Suppose 𝑉 is an 𝑛-dimensional complex vector

space, considered as an abelian complex Lie group. A lattice in 𝑉 is a discrete
additive subgroup Λ ⊆ 𝑉 generated by 2𝑛 vectors 𝑣1 , … , 𝑣2𝑛 that are linearly inde-
pendent over ℝ. Corollary 1.17 shows that 𝑉 /Λ is an 𝑛-dimensional complex Lie
group, called a complex torus. When 𝑛 = 0, it is just a single point. When 𝑛 > 0,
the real-linear isomorphism 𝐴 ∶ ℝ2𝑛 → 𝑉 given by 𝐴(𝑥1 , … , 𝑥2𝑛 ) = 𝑥𝑗 𝑣𝑗 de-
scends to a diffeomorphism from ℝ2𝑛 /ℤ2𝑛 to 𝑉 /Λ; since ℝ2𝑛 /ℤ2𝑛 is diffeomorphic
to the 2𝑛-torus 𝕊1 × ⋯ × 𝕊1 , so is 𝑉 /Λ. Thus the complex tori defined by different
lattices are all diffeomorphic to each other. They are typically not biholomorphic,
however; see Problem 1-4 for an example. //
Example 1.19 (Hopf Manifolds). Let 𝜆 = (𝜆1 , … , 𝜆𝑛 ) be an ordered 𝑛-tuple of
real numbers with 0 < 𝜆𝑗 < 1, and define an action of ℤ on ℂ𝑛 ∖ {0} by 𝑘 ⋅
𝑧 = ((𝜆1 )𝑘 𝑧1 , … , (𝜆𝑛 )𝑘 𝑧𝑛 ). This action is holomorphic, free, and proper, so the
quotient 𝐻𝜆 = (ℂ𝑛 ∖ {0})/ℤ is an 𝑛-dimensional complex manifold called a Hopf
manifold. Regarding 𝕊2𝑛−1 as the set of unit vectors in ℂ𝑛 , we define a smooth map
𝐴 ∶ 𝕊2𝑛−1 ×ℝ → ℂ𝑛 ∖{0} by 𝐴(𝑧, 𝑡) = ((𝜆1 )𝑡 𝑧1 , … , (𝜆𝑛 )𝑡 𝑧𝑛 ); if 𝜋 ∶ ℂ𝑛 ∖ {0} → 𝐻𝜆
is the quotient map, one can check that 𝜋 ∘ 𝐴 makes the same identifications as the
quotient map from 𝕊2𝑛−1 ×ℝ to 𝕊2𝑛−1 ×(ℝ/ℤ) ≈ 𝕊2𝑛−1 × 𝕊1 , so all Hopf manifolds
are diffeomorphic to 𝕊2𝑛−1 × 𝕊1 . //
Example 1.20 (Iwasawa Manifolds). Consider the subgroup 𝐺 ⊆ GL(3, ℂ) con-
sisting of matrices of the form

⎛1 𝑧 𝑧 ⎞
1 3
⎜0 1 𝑧2 ⎟
⎜ ⎟
⎝0 0 1 ⎠
12 1. The Basics

for 𝑧1 , 𝑧2 , 𝑧3 ∈ ℂ. It is a complex Lie group, biholomorphic to ℂ3 , with multipli-

(𝑧1 , 𝑧2 , 𝑧3 ) ⋅ (𝑤1 , 𝑤2 , 𝑤3 ) = (𝑧1 + 𝑤1 , 𝑧2 + 𝑤2 , 𝑧3 + 𝑤3 + 𝑧1 𝑤2 ).

cation given by

For a discrete subgroup Γ ⊆ 𝐺, the left coset space 𝐺/Γ is a complex 3-manifold
by Corollary 1.17. An Iwasawa manifold is a left coset space of the form 𝐺/Γ for
a discrete subgroup Γ that is cocompact, meaning that 𝐺/Γ is compact. (Some au-
thors use quotients by left Γ-actions in their definitions, corresponding to right coset
spaces; group inversion in 𝐺 induces a biholomorphism between the left and right

Iwasawa manifold, obtained by taking Γ to be the subgroup consisting of matrices

coset spaces, so there is no real difference.) The simplest example is the standard

in which 𝑧1 , 𝑧2 , 𝑧3 are Gaussian integers, that is, complex numbers of the form
𝑚 + 𝑛𝑖 for 𝑚, 𝑛 ∈ ℤ. It is cocompact by the result of Problem 1-1. //

Some Complex Analysis

In this book, we assume you are familiar with basic undergraduate-level complex
analysis in one variable; if your complex analysis is rusty, this would be a good time
to review. (Some suggested texts are listed in the Preface.)

𝑊 ⊆ ℂ is an open subset and 𝑓 ∶ 𝑊 → ℂ is a function, then 𝑓 is said to be

Recall the definition of a holomorphic function of one complex variable: if

holomorphic if it has a complex derivative at each point 𝑝 ∈ 𝑊 , defined by

𝑓 (𝑎 + ℎ) − 𝑓 (𝑎)
𝑓 ′ (𝑎) = lim .
ℎ→0 ℎ
Holomorphic functions are sometimes called complex-analytic, or just analytic if
there can be no confusion with real-analytic functions.

In these statements, 𝑊 represents an arbitrary open subset of ℂ and 𝑓 ∶ 𝑊 → ℂ

For convenience, let us recall some basic facts from the one-variable theory.

is an arbitrary holomorphic function. We write the standard coordinate on ℂ as

𝑧 = 𝑥 + 𝑖𝑦.
• CAUCHY INTEGRAL FORMULA: If 𝑎 ∈ 𝑊 and 𝑟 > 0 is chosen so that the
closed disk 𝐷𝑟 (𝑎) is contained in 𝑊 , then the following formula holds for
all 𝑧 in the open disk 𝐷𝑟 (𝑎):
1 ℎ(𝜁)
ℎ(𝑧) = 𝑑𝜁.
2𝜋𝑖 ∫ 𝜁 −𝑧
|𝜁−𝑎|=𝑟

• CAUCHY–RIEMANN EQUATIONS: The real and imaginary parts 𝑢 and 𝑣 of

𝑓 satisfy the equations
𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣
= , =− .
𝜕𝑥 𝜕𝑦 𝜕𝑦 𝜕𝑥
Some Complex Analysis 13

• POWER SERIES EXPANSION: For each 𝑎 ∈ 𝑊 , if the disk 𝐷𝑟 (𝑎) is con-

tained in 𝑊 , then 𝑓 |𝐷 is equal to a convergent series in powers of (𝑧 − 𝑎).
It has complex derivatives of all orders, which may be computed by dif-
ferentiating the series term-by-term.
• ZEROS ARE ISOLATED AND HAVE FINITE ORDER: If 𝑓 (𝑎) = 0 for some
𝑎 ∈ 𝑊 and 𝑓 is not identically zero, then there is a disk 𝐷𝑟 (𝑎) ⊆ 𝑊 such
that 𝑓 (𝑧) ≠ 0 for 𝑧 ∈ 𝐷𝑟 (𝑎) ∖ {𝑎}; and there is a positive integer 𝑚 (called
the order or multiplicity of the zero), such that 𝑓 (𝑧) = (𝑧 − 𝑎)𝑚 ℎ(𝑧) for
some holomorphic function ℎ that does not vanish at 𝑎. The order of a
zero is equal to the smallest integer 𝑚 such that 𝑓 (𝑚) (𝑎) ≠ 0. A zero of
order 1 is called a simple zero.
• MAXIMUM PRINCIPLE: If 𝑊 is connected and |𝑓 (𝑧)| attains a maximum
at a point 𝑧 ∈ 𝑊 , then 𝑓 is constant.
• LIOUVILLE’S THEOREM: If 𝑊 = ℂ and 𝑓 is bounded, then it is constant.
• RIEMANN’S REMOVABLE SINGULARITY THEOREM: If 𝑊 = 𝑊 ̃ ∖ {𝑎} for
some open set 𝑊 ̃ and some point 𝑎 ∈ 𝑊 ̃ , and 𝑓 is bounded, then 𝑓
̃.
extends to a holomorphic function on all of 𝑊

For our study of complex manifolds, we need to extend some of the results of
the one-variable theory to functions of several complex variables. Many of these re-
sults will look familiar, but some properties of holomorphic functions are decidedly
different in higher dimensions.

ables. Suppose 𝑈 ⊆ ℂ𝑛 is an open subset and 𝑓 ∶ 𝑈 → ℂ. For 𝑝 = (𝑝1 , … , 𝑝𝑛 ) ∈ 𝑈

We begin with the official definition of holomorphic functions of several vari-

and 𝑗 ∈ {1, … , 𝑛}, we say 𝑓 has a complex partial derivative at 𝒑 with respect to
𝒛𝒋 if the following limit exists:

𝜕𝑓 𝑓 (𝑝1 , … , 𝑝𝑗 + ℎ, … , 𝑝𝑛 ) − 𝑓 (𝑝1 , … , 𝑝𝑛 )
(𝑝) = ,
𝜕𝑧𝑗 ℎ→0 ℎ
(1.2) lim

where the limit is taken over all ℎ in some punctured disk centered at the origin in
ℂ. Such a function is said to be holomorphic if it is continuous and has a complex
partial derivative with respect to each variable 𝑧1 , … , 𝑧𝑛 at each point of 𝑈 . More
generally, a vector-valued function 𝐹 ∶ 𝑈 → ℂ𝑘 is said to be holomorphic if each
of its component functions is holomorphic.
Our definition of holomorphic functions is essentially the same as the one-
variable definition, except in that case the assumption of continuity is not needed
because a simple argument shows that continuity follows from the existence of a
complex derivative. It is worth noting, in fact, that the continuity assumption is ac-
tually not needed in higher dimensions either: the German mathematician Friedrich
14 1. The Basics

tives at every point of an open subset of ℂ𝑛 is automatically continuous. That proof

Hartogs proved in 1906 [Har06] that a function that has complex partial deriva-

(which can be found in [Kra01, Section 2.4]) is difficult, though, so it is much more
convenient simply to assume continuity as part of our definition.
In one complex variable, there are several equivalent ways to characterize holo-
morphic functions: having a complex derivative everywhere, or having continuous
partial derivatives that satisfy the Cauchy–Riemann equations, or being the sum
of a convergent power series in a neighborhood of each point. There are similar
equivalent characterizations for holomorphic functions of several variables.

Theorem 1.21. Let 𝑈 ⊆ ℂ𝑛 be open and 𝑓 ∶ 𝑈 → ℂ. The following are equivalent.

(a) 𝑓 is holomorphic (i.e., it is continuous and has a complex partial deriva-
tive with respect to each variable at each point of 𝑈 ).
(b) 𝑓 is smooth and satisfies the following Cauchy–Riemann equations:

𝜕𝑢 𝜕𝑣 𝜕𝑢 𝜕𝑣
= 𝑗, = − 𝑗, 𝑗 = 1, … , 𝑛,
𝜕𝑥 𝑗 𝜕𝑦 𝜕𝑦𝑗 𝜕𝑥

where 𝑧𝑗 = 𝑥𝑗 + 𝑖𝑦𝑗 and 𝑓 (𝑧) = 𝑢(𝑧) + 𝑖𝑣(𝑧).

(c) For each 𝑝 = (𝑝1 , … , 𝑝𝑛 ) ∈ 𝑈 , there exists a neighborhood of 𝑝 in 𝑈 on
which 𝑓 is equal to the sum of an absolutely convergent power series of
the form
∞
𝑓 (𝑧) = 𝑎𝑘1 …𝑘𝑛 (𝑧1 − 𝑝1 )𝑘1 ⋯ (𝑧𝑛 − 𝑝𝑛 )𝑘𝑛 .
∑
(1.3)
𝑘1 ,…,𝑘𝑛 =0

Remarks.
• In the decomposition 𝑓 (𝑧) = 𝑢(𝑧) + 𝑖𝑣(𝑧) in part (b), it is understood that
𝑢(𝑧) and 𝑣(𝑧) are real. The same applies to 𝑧𝑗 = 𝑥𝑗 +𝑖𝑦𝑗 and everywhere in
the book when we write such a decomposition, unless otherwise specified.
• In (c), the reason we insist on absolute convergence is that a sum over
multiple indices can be ordered in various ways, and absolute convergence
ensures that the ordering of terms does not matter.

Proof. We will prove (a) ⇔ (b) and (a) ⇔ (c).

Suppose 𝑓 satisfies (a). Because 𝑓 is holomorphic in each variable separately,

respect to each variable. To show that it is smooth, given 𝑝 ∈ 𝑈 , choose 𝑟 > 0

the one-variable theory shows that it satisfies the Cauchy–Riemann equations with

such that the closed polydisk 𝐷𝑛𝑟 (𝑝) is contained in 𝑈 . Because 𝑓 is holomorphic
in each variable separately, we can apply the single-variable version of the Cauchy
Some Complex Analysis 15

integral formula repeatedly to obtain the following for all 𝑧 ∈ 𝐷𝑟𝑛 (𝑝):
𝑓 (𝑧1 , … , 𝑧𝑛 )
1 𝑓 (𝑧1 , … , 𝑧𝑛−1 , 𝜁 𝑛 ) 𝑛
= 𝑑𝜁
2𝜋𝑖 ∫ 𝜁 𝑛 − 𝑧𝑛
|𝜁 𝑛 −𝑝𝑛 |=𝑟

1 𝑓 (𝑧1 , … , 𝜁 𝑛−1 , 𝜁 𝑛 )
= 𝑑𝜁 𝑛−1 𝑑𝜁 𝑛
(1.4) (2𝜋𝑖)2 ∫ ∫ (𝜁 𝑛 − 𝑧𝑛 )(𝜁 𝑛−1 − 𝑧𝑛−1 )
|𝜁 𝑛 −𝑝𝑛 |=𝑟 |𝜁 𝑛−1 −𝑝𝑛−1 |=𝑟

⋮
1 𝑓 (𝜁 1 , … , 𝜁 𝑛 )
= … 𝑑𝜁 1 ⋯ 𝑑𝜁 𝑛 .
(2𝜋𝑖)𝑛 ∫ ∫ (𝜁 𝑛 − 𝑧𝑛 ) ⋯ (𝜁 1 − 𝑧1 )
|𝜁 𝑛 −𝑝𝑛 |=𝑟 |𝜁 1 −𝑝1 |=𝑟

variables and smooth as a function of (the real and imaginary parts of) 𝑧1 , … , 𝑧𝑛 ,
Since the domain of integration is compact and the integrand is continuous in all

we can differentiate under the integral sign as often as we like with respect to 𝑥𝑗
and 𝑦𝑗 to conclude that 𝑓 is smooth. This proves (b).
To prove that 𝑓 also satisfies (c), note that
1 1 1 1
= 𝑗 = 𝑗 ,
𝜁𝑗 −𝑧 𝑗 (𝜁 − 𝑝 ) − (𝑧 − 𝑝 ) 𝜁 − 𝑝𝑗
𝑗 𝑗 𝑗
𝑧𝑗 − 𝑝 𝑗
1−( 𝑗
𝜁 − 𝑝𝑗 )
and since |𝑧𝑗 − 𝑝𝑗 |/|𝜁 𝑗 − 𝑝𝑗 | < 1 on the domain of integration in (1.4), we can
expand the last fraction on the right in a power series to obtain

𝑧𝑗 − 𝑝 𝑗 𝑘
∞
1 1
= ,
𝜁 𝑗 − 𝑧𝑗 ∑ ( 𝜁 𝑗 − 𝑝𝑗 )
𝜁 𝑗 − 𝑝𝑗 𝑘=0

which converges uniformly and absolutely for 𝑧𝑗 in any closed disk 𝐷𝑟′ (𝑝𝑗 ) with 0 <
𝑟′ < 𝑟 by comparison with the geometric series ∑𝑘 (𝑟′ /𝑟)𝑘 . Inserting this formula
for each variable into (1.4), we conclude that 𝑓 satisfies (1.3) with coefficients
𝑓 (𝜁 1 , … , 𝜁 𝑛 )
𝑎𝑘1 …𝑘𝑛 = … 𝑑𝜁 1 ⋯ 𝑑𝜁 𝑛 .
∫ ∫ (𝜁 𝑛 − 𝑝𝑛 )𝑘𝑛 +1 ⋯ (𝜁 1 − 𝑝1 )𝑘1 +1
|𝜁 𝑛 −𝑝𝑛 |=𝑟 |𝜁 1 −𝑝1 |=𝑟

This completes the proof that (a) ⇒ (b) and (a) ⇒ (c).
Conversely, if 𝑓 satisfies (b), then it is certainly continuous, and the one-
variable theory implies that it has a complex derivative with respect to each variable,

Finally, assume 𝑓 satisfies (c), and let 𝑝 ∈ 𝑈 be arbitrary. There is some closed
so it also satisfies (a).

polydisk 𝐷𝑛𝑟 (𝑝) contained in 𝑈 and centered at 𝑝 on which the series converges
absolutely. Because the series converges at 𝑧0 = (𝑝1 + 𝑟, … , 𝑝𝑛 + 𝑟), the terms in
16 1. The Basics

the series for 𝑓 (𝑧0 ) are all uniformly bounded, which means there is a constant 𝐶
such that
|𝑎𝑘1 …𝑘𝑛 |𝑟𝑘1 ⋯ 𝑟𝑘𝑛 ≤ 𝐶.
On a polydisk 𝐷𝑛𝑟′ (𝑝) for any 0 < 𝑟′ < 𝑟, the terms of the series satisfy the following
bound:
∞

|𝑎𝑘1 …𝑘𝑛 (𝑧 − 𝑝 ) ⋯ (𝑧 − 𝑝 ) |
1 1 𝑘1 𝑛 𝑛 𝑘𝑛
∑
(1.5)
𝑘1 ,…,𝑘𝑛 =0
∞
𝑟′ 1 𝑟′ 𝑛
𝑘 𝑘
≤ 𝐶 ⋯ ,
∑ (𝑟) (𝑟)
𝑘1 ,…,𝑘𝑛 =0

the series for 𝑓 converges uniformly and absolutely on 𝐷𝑛𝑟′ (𝑝) by the Weierstrass
and the series on the right is an iterated convergent geometric series. Therefore,

M-test, so 𝑓 is continuous there, and in particular at 𝑝. If we fix all the variables

but 𝑧𝑗 , we obtain a convergent power series in 𝑧𝑗 , which is therefore holomorphic
in 𝑧𝑗 by the one-variable theory, thus proving (a). □

Next we enumerate the basic properties of holomorphic functions that we will

use throughout the book.

Suppose 𝑍 ⊆ ℂ𝑚 and 𝑊 ⊆ ℂ𝑛 are open subsets and 𝑓 ∶ 𝑍 → 𝑊 and 𝑔 ∶ 𝑊 → ℂ𝑘

Proposition 1.22 (Compositions of Holomorphic Functions are Holomorphic).

are holomorphic functions. Then 𝑔 ∘ 𝑓 ∶ 𝑍 → ℂ𝑘 is holomorphic.

Proof. Certainly 𝑔 ∘ 𝑓 is smooth, so we just need to check that it satisfies the

Cauchy–Riemann equations. Let us write the variables in 𝑍 as 𝑧𝑗 = 𝑥𝑗 + 𝑖𝑦𝑗 ,
those in 𝑊 as 𝑤𝑗 = 𝑢𝑗 + 𝑖𝑣𝑗 , and the component functions of 𝑓 and 𝑔 as
𝑓 𝑘 (𝑧) = 𝑈 𝑘 (𝑧) + 𝑖𝑉 𝑘 (𝑧), 𝑔 𝑙 (𝑤) = 𝐴𝑙 (𝑤) + 𝑖𝐵 𝑙 (𝑤). Applying the real-variable
chain rule (and using the summation convention), we find

𝜕(𝐴𝑙 ∘ 𝑓 ) 𝜕(𝐵 𝑙 ∘ 𝑓 ) 𝜕𝐴𝑙 𝜕𝑈 𝑘 𝜕𝐴𝑙 𝜕𝑉 𝑘 𝜕𝐵 𝑙 𝜕𝑈 𝑘 𝜕𝐵 𝑙 𝜕𝑉 𝑘

− = + − − .
𝜕𝑥𝑗 𝜕𝑦𝑗 𝜕𝑢𝑘 𝜕𝑥𝑗 𝜕𝑣𝑘 𝜕𝑥𝑗 𝜕𝑢𝑘 𝜕𝑦𝑗 𝜕𝑣𝑘 𝜕𝑦𝑗

Using the Cauchy–Riemann equations for 𝑔 to replace 𝜕𝐴𝑙 /𝜕𝑢𝑘 by 𝜕𝐵 𝑙 /𝜕𝑣𝑘 and
𝜕𝐴𝑙 /𝜕𝑣𝑘 by −𝜕𝐵 𝑙 /𝜕𝑢𝑘 and then applying the Cauchy–Riemann equations for 𝑓 , we
see that this expression is identically zero. A similar computation shows that the
composition also satisfies the other set of Cauchy–Riemann equations. □

Proposition 1.23. Suppose 𝑓 , 𝑔 ∶ 𝑈 → ℂ are holomorphic functions on an open

subset 𝑈 ⊆ ℂ𝑛 . Then 𝑓 + 𝑔, 𝑓 − 𝑔, and 𝑓 𝑔 are holomorphic on 𝑈 , and 𝑓 /𝑔 is
holomorphic on 𝑈 ∖ 𝑔 −1 (0).
Some Complex Analysis 17

► Exercise 1.24. Prove this proposition.

polynomial functions of 𝑧1 , … , 𝑧𝑛 are holomorphic on ℂ𝑛 , and all rational func-

It follows easily from the two preceding propositions, for example, that all

tions (quotients of polynomials) are holomorphic wherever their denominators are

nonzero.

ables to those with respect to real variables. If 𝑓 = 𝑢 + 𝑖𝑣 is a complex-valued func-

Our next proposition relates partial derivatives with respect to complex vari-

tion, the notation 𝜕𝑓 /𝜕𝑥𝑗 denotes the complex-valued function 𝜕𝑢/𝜕𝑥𝑗 + 𝑖𝜕𝑣/𝜕𝑥𝑗 ,
and similarly with 𝑦𝑗 derivatives.

Proposition 1.25. Suppose 𝑈 is an open subset of ℂ𝑛 and 𝑓 ∶ 𝑈 → ℂ is holomor-

phic. Writing 𝑧𝑗 = 𝑥𝑗 + 𝑖𝑦𝑗 for 𝑗 = 1, … , 𝑛, we have

𝜕𝑓 𝜕𝑓 1 𝜕𝑓
= 𝑗 = .
𝜕𝑧 𝑗 𝜕𝑥 𝑖 𝜕𝑦𝑗
(1.6)

Proof. Note that the existence of the limit in (1.2) as ℎ approaches zero through all
complex values implies that we obtain the same limit if we restrict ℎ to approach
zero through real values only or imaginary values only. Thus for any 𝑝 ∈ 𝑈 ,

𝜕𝑓 𝑓 (𝑝1 , … , 𝑝𝑗 + ℎ, … , 𝑝𝑛 ) − 𝑓 (𝑝1 , … , 𝑝𝑛 ) 𝜕𝑓
(𝑝) = = 𝑗 (𝑝),
𝜕𝑧𝑗 ℎ→0 ℎ 𝜕𝑥
lim
ℎ∈ℝ

𝜕𝑓 𝑓 (𝑝1 , … , 𝑝𝑗 + 𝑖𝑘, … , 𝑝𝑛 ) − 𝑓 (𝑝1 , … , 𝑝𝑛 ) 1 𝜕𝑓

(𝑝) = = (𝑝).
𝜕𝑧𝑗 𝑘→0 𝑖𝑘 𝑖 𝜕𝑦𝑗
lim □
𝑘∈ℝ

Next we establish some important properties of multivariable power series.

Proposition 1.26. Suppose 𝑓 is a holomorphic function given by an absolutely

convergent power series of the form (1.3) on a polydisk 𝐷𝑟𝑛 (𝑝) ⊆ ℂ𝑛 . The complex
partial derivatives of 𝑓 of all orders exist and are given by absolutely convergent
power series on the same polydisk, which can be computed by differentiating the
series term by term.

Proof. For any 0 < 𝑟′ < 𝑟1 < 𝑟, the series converges absolutely on 𝐷𝑛𝑟1 (𝑝), and

on 𝐷𝑟𝑛′ (𝑝). Note that the complex derivative 𝜕𝑓 /𝜕𝑧𝑗 is equal to the real partial
thus the proof of Theorem 1.21 shows that it converges uniformly and absolutely

derivative 𝜕𝑓 /𝜕𝑥𝑗 by Proposition 1.25. A standard result in real analysis [Rud76,

respect to 𝑥𝑗 on 𝐷𝑟𝑛′ (𝑝) provided the differentiated series converges uniformly there.
Thm. 7.17] shows that we can differentiate the power series term by term with
18 1. The Basics

With notation as in (1.5), the differentiated series satisfies

∞
𝜕
| 𝜕𝑥𝑗 (𝑎𝑘1 …𝑘𝑛 (𝑧 − 𝑝 ) ⋯ (𝑧 − 𝑝 ) )|
1 1 𝑘1 𝑛 𝑛 𝑘𝑛
∑
𝑘1 ,…,𝑘𝑛 =0
∞
= |𝑎𝑘1 …𝑘𝑛 (𝑧 − 𝑝 ) ⋯ 𝑘𝑗 (𝑧 − 𝑝 )
1 1 𝑘1 𝑗 𝑗 𝑘𝑗 −1
⋯ (𝑧𝑛 − 𝑝𝑛 )𝑘𝑛 |
∑
𝑘1 ,…,𝑘𝑛 =0
∞
𝑟′ 1 𝑟′ 𝑟′ 𝑛
𝑘 𝑘𝑗 −1 𝑘
≤ 𝐶 ⋯ 𝑘𝑗 ⋯ .
∑ ( 𝑟1 ) ( 𝑟1 ) ( 𝑟1 )
𝑘1 ,…,𝑘 =0 𝑛

The last expression is an iterated sum in which 𝑛 − 1 of the sums are convergent
geometric series, while the 𝑗th one is the series ∑𝑘 𝑘𝑥𝑘−1 , which converges ab-
solutely for |𝑥| < 1 by the ratio test. Thus we may apply the Weierstrass M-test

on 𝐷𝑟𝑛′ (𝑝), and therefore is equal to the derivative of 𝑓 there. Since every point in
again to conclude that the differentiated series converges uniformly and absolutely

𝐷𝑟𝑛 (𝑝) lies in 𝐷𝑟𝑛′ (𝑝) for some 0 < 𝑟′ < 𝑟1 < 𝑟, it follows that 𝜕𝑓 /𝜕𝑧𝑗 is equal to the
sum of the differentiated series on all of 𝐷𝑛𝑟 (𝑝). It then follows by induction that the
same is true of all higher complex derivatives. □

Corollary 1.27. If 𝑓 is a holomorphic function given by an absolutely convergent

power series of the form (1.3) on a polydisk 𝐷𝑟𝑛 (𝑝) ⊆ ℂ𝑛 , then the power series is
given explicitly by the following formula, called the Taylor series of 𝒇 centered at
𝒑:

𝜕 𝑘1 +⋯+𝑘𝑛 𝑓 (𝑝)
∞
1
𝑓 (𝑧) = (𝑧1 − 𝑝1 )𝑘1 ⋯ (𝑧𝑛 − 𝑝𝑛 )𝑘𝑛 .
∑ 𝑘1 ! ⋯ 𝑘𝑛 ! (𝜕𝑧1 )𝑘1 ⋯ (𝜕𝑧𝑛 )𝑘𝑛
𝑘1 ,…,𝑘 =0 𝑛

Proof. Just differentiate (1.3) repeatedly term-by-term and evaluate at 𝑧 = 𝑝 to

determine the coefficients 𝑎𝑘1 …𝑘𝑛 . □

Proposition 1.28 (Identity Theorem). Suppose 𝑊 ⊆ ℂ𝑛 is a connected open

subset, and 𝑓 , 𝑔 ∶ 𝑊 → ℂ are holomorphic functions that agree on a nonempty
open subset of 𝑊 . Then 𝑓 ≡ 𝑔 on 𝑊 .

Proof. Set ℎ = 𝑓 − 𝑔, so ℎ ≡ 0 on a nonempty open subset 𝑈0 ⊆ 𝑊 . Let

𝑈 = {𝑝 ∈ 𝑊 ∶ ℎ and its complex partial derivatives of all orders vanish at 𝑝}.
Then 𝑈 is nonempty because 𝑈0 ⊆ 𝑈 . We will show that it is open and closed in
𝑊 , which implies by connectivity that it is all of 𝑊 .
Suppose 𝑝 ∈ 𝑈 . Then ℎ is equal to a convergent power series in a neighborhood
of 𝑝, and Corollary 1.27 shows that every term in the series is zero. Thus 𝑈 is open
in 𝑊 .
Some Complex Analysis 19

Now suppose 𝑝 ∈ 𝑊 is a limit point of 𝑈 . There is a sequence of points 𝑝𝑗 ∈ 𝑈

converging to 𝑝, and the hypothesis implies that all partial derivatives of ℎ vanish
at each 𝑝𝑗 . Thus by continuity, they also vanish at 𝑝, showing that 𝑝 ∈ 𝑈 . Thus 𝑈
is closed in 𝑊 . □

Corollary 1.29 (Identity Theorem for Manifolds). Suppose 𝑀 and 𝑁 are com-
plex manifolds with 𝑀 connected, and 𝑓 , 𝑔 ∶ 𝑀 → 𝑁 are holomorphic maps that
agree on a nonempty open subset of 𝑀. Then 𝑓 ≡ 𝑔 on 𝑀.

Proof. Proposition 1.28 applied to local coordinate representations of 𝑓 and 𝑔

shows that the set of points where 𝑓 and 𝑔 agree along with their partial deriva-
tives of all orders is both open and closed in 𝑀, hence all of 𝑀. □

fined on all of ℂ𝑛 and bounded is constant.

Proposition 1.30 (Liouville’s Theorem). Every holomorphic function that is de-

Proof. Suppose 𝑓 ∶ ℂ𝑛 → ℂ is holomorphic and bounded. Given any point 𝑧 ∈

ℂ𝑛 , the function 𝑔(𝜁) = 𝑓 (𝜁𝑧) is a bounded holomorphic function defined on all of
ℂ, so it is constant by the one-variable version of Liouville’s theorem. In particular,
this means 𝑓 (𝑧) = 𝑓 (0). Since 𝑧 is arbitrary, this shows 𝑓 is constant. □

Liouville’s theorem allows us to give our first example of two complex mani-
folds that are diffeomorphic but not biholomorphic.
Example 1.31 (The Unit Ball is Not Biholomorphic to ℂ𝒏 ). We know that 𝔹2𝑛
and ℂ𝑛 are diffeomorphic (see [LeeSM, Example 2.14]). But if 𝐹 ∶ ℂ𝑛 → 𝔹2𝑛 is

function on ℂ𝑛 and therefore constant. Thus there is no biholomorphism between

any holomorphic map, each of its coefficient functions is a bounded holomorphic

𝔹2𝑛 and ℂ𝑛 . //
Proposition 1.32 (The Maximum Principle). Suppose 𝑓 ∶ 𝑈 → ℂ is a holomor-
phic function on a connected open set 𝑈 ⊆ ℂ𝑛 . If |𝑓 (𝑧)| attains a maximum value
at some point in 𝑈 , then 𝑓 is constant.

Proof. Suppose |𝑓 (𝑧)| attains a maximum value at 𝑧0 ∈ 𝑈 . Let 𝑐 = 𝑓 (𝑧0 ), and

set 𝑊 = {𝑧 ∈ 𝑈 ∶ 𝑓 (𝑧) = 𝑐}. Then 𝑊 is nonempty because 𝑧0 ∈ 𝑊 , and it is
closed in 𝑈 by continuity. Given 𝑧1 ∈ 𝑊 , choose 𝜀 > 0 such that the ball 𝐵𝜀 (𝑧1 )
is contained in 𝑈 . For each 𝑤 ∈ ℂ𝑛 with |𝑤| = 1, the function 𝑔(𝜁) = 𝑓 (𝑧1 + 𝜁𝑤)
is holomorphic on the disk 𝐷𝜀 (0) ⊆ ℂ and achieves its maximum modulus at 𝜁 =
0. By the one-variable maximum principle, therefore, 𝑔 is constant. Since 𝑤 is
arbitrary, this shows 𝑓 is constant on 𝐵𝜀 (𝑧1 ). Thus 𝑊 is open, and by connectivity
it is all of 𝑈 . □

This result too has an immediate, and somewhat surprising, application to com-
plex manifolds.
20 1. The Basics

Corollary 1.33. Let 𝑀 be a connected compact complex manifold. Then every

globally defined holomorphic function from 𝑀 to ℂ is constant.

Proof. Suppose 𝑓 ∈ 𝒪(𝑀). By compactness, the continuous function |𝑓 | attains

a maximum value at a point 𝑧0 ∈ 𝑀. In a holomorphic coordinate ball centered
at 𝑧0 , the coordinate representation of 𝑓 is a holomorphic function on an open ball
in ℂ𝑛 that attains its maximum modulus at the origin, so it is constant on the entire
coordinate domain. Thus by the identity theorem, it is constant on all of 𝑀. □

One of the most striking features of holomorphic functions is described in the

next proposition, which shows in particular that uniform limits of holomorphic
functions are holomorphic. It is worth noting that the analogous result for smooth
functions, or even real-analytic functions, is not true.
Proposition 1.34. Suppose 𝑈 ⊆ ℂ𝑛 is open and 𝑓𝑘 ∶ 𝑈 → ℂ is a sequence of holo-
morphic functions that converge uniformly on compact subsets of 𝑈 to a function
𝑓 ∶ 𝑈 → ℂ. Then 𝑓 is holomorphic.

Proof. Given 𝑝 ∈ 𝑈 , choose 𝑟 > 0 such that 𝐷𝑛𝑟 (𝑝) ⊆ 𝑈 . For all 𝑧 ∈ 𝐷𝑟𝑛 (𝑝), we
can apply the Cauchy integral formula to 𝑓𝑘 , and uniform convergence guarantees
that
1 𝑓𝑘 (𝜁 1 , … , 𝜁 𝑛 )
𝑓 (𝑧) = lim … 𝑑𝜁 1 ⋯ 𝑑𝜁 𝑛
𝑘→∞ (2𝜋𝑖)𝑛 ∫ ∫ (𝜁 𝑛 − 𝑧𝑛 ) ⋯ (𝜁 1 − 𝑧1 )
|𝜁 𝑛 −𝑝𝑛 |=𝑟 |𝜁 1 −𝑝1 |=𝑟

1 𝑓 (𝜁 1 , … , 𝜁 𝑛 )
= … 𝑑𝜁 1 ⋯ 𝑑𝜁 𝑛 .
(2𝜋𝑖)𝑛 ∫ ∫ (𝜁 𝑛 − 𝑧𝑛 ) ⋯ (𝜁 1 − 𝑧1 )
|𝜁 𝑛 −𝑝𝑛 |=𝑟 |𝜁 1 −𝑝1 |=𝑟

𝑧1 , … , 𝑧𝑛 , so we can differentiate under the integral sign with respect to 𝑥𝑗 and

The integrand in the last expression is continuous in all variables and smooth in

𝑦𝑗 as many times as we like to conclude that 𝑓 is smooth. In particular, since the

integrand is holomorphic in 𝑧1 , … , 𝑧𝑛 , we see that 𝑓 satisfies the Cauchy–Riemann
equations on 𝐷𝑛𝑟 (𝑝). □

Our next result is a little less elementary, so its one-variable analogue is not
always covered in undergraduate complex analysis texts. We will use it only once,
when we study sections of holomorphic vector bundles (Thm. 3.13).
Proposition 1.35 (Montel’s Theorem). Suppose 𝑈 ⊆ ℂ𝑛 is open and 𝑓𝑘 ∶ 𝑈 → ℂ

is some 𝐶 > 0 such that |𝑓𝑘 (𝑧)| < 𝐶 for all 𝑘 ≥ 1 and all 𝑧 ∈ 𝑈 . Then there
is a sequence of holomorphic functions that are uniformly bounded, meaning there

is a subsequence {𝑓𝑘𝑗 }∞
𝑗=1 that converges uniformly on compact subsets of 𝑈 to a
holomorphic function defined on all of 𝑈 .
Some Complex Analysis 21

Proof. For any closed polydisk 𝐷𝑛𝑟 (𝑝) ⊆ 𝑈 , we can use Cauchy’s formula to write
each 𝑓𝑘 on 𝐷𝑛𝑟 (𝑝) in the form

1 𝑓𝑘 (𝜁 1 , … , 𝜁 𝑛 )
𝑓𝑘 (𝑧) = … 𝑑𝜁 1 ⋯ 𝑑𝜁 𝑛 .
(2𝜋𝑖)𝑛 ∫ ∫ (𝜁 𝑛 − 𝑧𝑛 ) ⋯ (𝜁 1 − 𝑧1 )
|𝜁 𝑛 −𝑝𝑛 |=𝑟 |𝜁 1 −𝑝1 |=𝑟

Differentiating under the integral sign, we obtain

𝜕𝑓𝑘 (𝑧) 1 𝑓𝑘 (𝜁 1 , … , 𝜁 𝑛 )
= … 𝑑𝜁 1 ⋯ 𝑑𝜁 𝑛 .
𝜕𝑥𝑗 (2𝜋𝑖)𝑛 ∫ ∫ (𝜁 𝑛 − 𝑧𝑛 ) ⋯ (𝜁 𝑗 − 𝑧𝑗 )2 ⋯ (𝜁 1 − 𝑧1 )
A simple computation shows that a contour integral over a circle 𝑐 of radius 𝑟 sat-
isfies |∫𝑐 ℎ(𝜁)𝑑𝜁 | ≤ 2𝜋𝑟 sup𝑐 |ℎ|. Applying this in turn to each contour integral in

𝜕𝑓𝑘 (𝑧)
the above formula gives
𝐶
≤ .
| 𝜕𝑥𝑗 | 𝑟
This shows that the partial derivatives of 𝑓𝑘 with respect to 𝑥1 , … , 𝑥𝑛 are uniformly
bounded on 𝐷𝑛𝑟 (𝑝), and then the Cauchy–Riemann equations show the same is true
of the derivatives with respect to 𝑦1 , … , 𝑦𝑛 . Therefore, each 𝑓𝑘 satisfies a Lipschitz
estimate of the form |𝑓𝑘 (𝑧1 ) − 𝑓𝑘 (𝑧2 )| ≤ (𝐶 ′ /𝑟)|𝑧1 − 𝑧2 | there. By continuity,
the same bound holds on the closed polydisk 𝐷𝑛𝑟 (𝑝). Thus the functions 𝑓𝑘 are
uniformly bounded and uniformly equicontinuous on 𝐷𝑛𝑟 (𝑝), so the Arzelà–Ascoli
theorem [Rud76, Thm. 7.25] guarantees that a subsequence {𝑓𝑘𝑗 }∞ 𝑗=1 converges

Every 𝑝 ∈ 𝑈 is contained in some polydisk 𝐷𝑟𝑛 (𝑝) such that 𝐷𝑛𝑟 (𝑝) ⊆ 𝑈 . The set
uniformly there.

of all such polydisks is an open cover of 𝑈 , and thus 𝑈 is covered by countably many
such polydisks. Let {𝑉𝑚 }∞ 𝑚=1 be such a countable cover. By the above argument,
we may choose a subsequence {𝑓1,𝑗 }∞ 𝑗=1 of the original sequence that converges
uniformly on 𝑉 1 . From that subsequence, we may choose a further subsequence
{𝑓2,𝑗 }∞𝑗=1 that also converges uniformly on 𝑉 2 . Continuing by induction, for each
𝑚 we get a subsequence {𝑓𝑚,𝑗 }∞ 𝑗=1 converging uniformly on 𝑉 1 ∪ ⋯ ∪ 𝑉 𝑚 , such
that the 𝑚th sequence is a subsequence of the (𝑚 − 1)st one. Finally, let {𝑓𝑘𝑗 }∞ 𝑗=1
be the diagonal subsequence 𝑓𝑘𝑗 = 𝑓𝑗,𝑗 . If 𝐾 ⊆ 𝑈 is any compact set, there is
some 𝑚 such that 𝐾 ⊆ 𝑉1 ∪ ⋯ ∪ 𝑉𝑚 . Since {𝑓𝑘𝑗 } is a subsequence of {𝑓𝑖,𝑗 } for
each 𝑖, it converges uniformly on 𝐾. By Proposition 1.34, the limit function is
holomorphic. □

So far, all these facts about holomorphic functions of several variables have
been straightforward generalizations of standard facts about holomorphic functions
of one variable. The next result, however, is radically different from anything in the
one-variable theory. It was proved by Friedrich Hartogs in 1906 [Har06].
22 1. The Basics

𝐷𝑅 (0) × {(𝑧2 , … , 𝑧𝑛 )} 𝑧2

|𝑧1 | = 𝑟1

Re 𝑧1

Im 𝑧1
𝐷𝑛𝑟 (0)

𝐷𝑅
𝑛
(0)

Figure 1.1. Proof of Hartogs’s extension theorem

Theorem 1.36 (Hartogs’s Extension Theorem). Let 𝑛 ≥ 2, and let Ω ⊆ ℂ𝑛 be

an open set of the form 𝐷𝑅
𝑛
(𝑝) ∖ 𝐷𝑛𝑟 (𝑝) for some 𝑝 ∈ ℂ𝑛 and 0 < 𝑟 < 𝑅. Every
holomorphic function 𝑓 ∶ Ω → ℂ has a unique extension to a holomorphic function
on all of 𝐷𝑅
𝑛
(𝑝).

Proof. After a translation, we may assume that 𝑝 = 0. Choose any 𝑟1 such that 𝑟 <
𝑟1 < 𝑅. As long as 𝑟 < |𝑧2 | < 𝑅, the function 𝑧1 ↦ 𝑓 (𝑧1 , … , 𝑧𝑛 ) is holomorphic
on the entire disk 𝐷𝑅 (0) ⊆ ℂ (see Fig. 1.1), so Cauchy’s formula shows that
1 𝑓 (𝜁, 𝑧2 , … , 𝑧𝑛 )
𝑓 (𝑧1 , … , 𝑧𝑛 ) = 𝑑𝜁.
2𝜋𝑖 ∫ 𝜁 − 𝑧1
|𝜁 |=𝑟1

But this formula actually makes sense for all (𝑧1 , … , 𝑧𝑛 ) ∈ 𝐷𝑟𝑛1 (0) because the
integration contour is contained in Ω in that case, and it defines a holomorphic
function 𝑓1 there by differentiation under the integral sign. Because 𝑓1 agrees with
𝑓 on the open subset of 𝐷𝑟𝑛1 (0) where 𝑟 < |𝑧2 | < 𝑟1 , the identity theorem shows
that it agrees on the entire connected set 𝐷𝑟𝑛1 (0) ∖ 𝐷𝑛𝑟 (0). Thus we can define a
holomorphic function on all of 𝐷𝑅 𝑛
(0) by letting it be equal to 𝑓 on Ω and to 𝑓1 on
𝐷𝑟1 (0). Uniqueness follows immediately from the identity theorem.
𝑛
□

This theorem is false in the case 𝑛 = 1, because there are many holomorphic
functions with isolated singularities, such as 1/𝑧 or 𝑒1/𝑧 , which are holomorphic
on annuli centered at a singular point but have no holomorphic extensions across
that point. Hartogs’s theorem implies that singularities of holomorphic functions
The Complexified Tangent and Cotangent Bundles 23

in two or more variables are never isolated. Moreover, it says something important
about zeros of holomorphic functions as well. In one complex variable, zeros of

function 𝑓 had an isolated zero at 𝑝 ∈ ℂ𝑛 with 𝑛 ≥ 2, then 1/𝑓 would have an

holomorphic functions of one variable are always isolated. But if a holomorphic

isolated singularity, which is impossible. Thus zeros of holomorphic functions of

more than one variable are never isolated either.

The Complexified Tangent and Cotangent Bundles

need for working with complex-valued functions. Writing such a function as 𝑓 =

Now we introduce some extensions to the theory of smooth manifolds that we will

𝑢 + 𝑖𝑣, we would like to express its differential as 𝑑𝑓 = 𝑑𝑢 + 𝑖𝑑𝑣. But this is not
an ordinary 1-form in the sense that the term is used in smooth manifold theory:
sections of a real vector bundle like the cotangent bundle can be multiplied by real

To make sense of this, we make the following definition. If 𝑉 is a real vector

numbers, but not by complex ones.

space, we define the complexification of 𝑽 , denoted by 𝑉ℂ , to be the vector space

𝑉 ⊕ 𝑉 with multiplication by complex numbers defined as follows:
(𝑎 + 𝑖𝑏)(𝑢, 𝑣) = (𝑎𝑢 − 𝑏𝑣, 𝑎𝑣 + 𝑏𝑢) for 𝑎 + 𝑖𝑏 ∈ ℂ.
Together with the usual addition in 𝑉 ⊕ 𝑉 , it turns 𝑉ℂ into a vector space over ℂ.
The map 𝑢 ↦ (𝑢, 0) is a real-linear isomorphism from 𝑉 onto the (real) subspace
𝑉 ⊕ {0} ⊆ 𝑉ℂ , and we typically identify 𝑉 with its image under this map, thus
considering 𝑉 itself to be a real-linear subspace of 𝑉ℂ . With this identification, we
can write (𝑢, 𝑣) = 𝑢 + 𝑖𝑣, and we can think of 𝑉ℂ as consisting of the set of all linear
combinations of elements of 𝑉 with complex coefficients.
If (𝑏1 , … , 𝑏𝑛 ) is any basis for 𝑉 (over ℝ), then ((𝑏1 , 0), … , (𝑏𝑛 , 0)) is a basis
for 𝑉ℂ over ℂ, which under our identification we can just write as (𝑏1 , … , 𝑏𝑛 ). It
follows that the complex dimension of 𝑉ℂ is the same as the real dimension of 𝑉 .
For example, the complexification of ℝ𝑛 can be naturally identified with ℂ𝑛 .
If 𝐿 ∶ 𝑉 → 𝑊 is a linear map between real vector spaces, it extends canoni-
cally to a complex-linear map 𝐿ℂ ∶ 𝑉ℂ → 𝑊ℂ , called the complexification of 𝑳,
satisfying 𝐿ℂ (𝑢 + 𝑖𝑣) = 𝐿(𝑢) + 𝑖𝐿(𝑣). In cases where it will not cause confusion,
we will often denote the complexification of a linear map 𝐿 by the same symbol 𝐿.

► Exercise 1.37. Show that the assignment 𝑉 ↦ 𝑉ℂ , 𝐿 ↦ 𝐿ℂ defines a co-

variant functor from the category of real vector spaces to the category of complex
ones.

The next exercise describes an alternative definition of the complexification.

► Exercise 1.38. Let 𝑉 be a real vector space. Give the space 𝑉 ⊗ℝ ℂ (the
abstract tensor product of 𝑉 and ℂ, considered as real vector spaces), the structure
24 1. The Basics

of a complex vector space with the usual addition and with scalar multiplication
defined by

𝑘 𝑘
𝛼( 𝑣𝑗 ⊗ 𝛽𝑗 ) = 𝑣𝑗 ⊗ (𝛼𝛽𝑗 ),
∑ ∑
𝑗=1 𝑗=1

for 𝑣𝑗 ∈ 𝑉 and 𝛼, 𝛽𝑗 ∈ ℂ. Show that this turns 𝑉 ⊗ℝ ℂ into a complex vector

space, which is canonically isomorphic to 𝑉ℂ via the map (𝑢, 𝑣) ↦ 𝑢 ⊗ 1 + 𝑣 ⊗ 𝑖.

► Exercise 1.39. Suppose 𝑉 is a real vector space.

(a) Given 𝑤 = (𝑢, 𝑣) ∈ 𝑉ℂ , define the conjugate of 𝒘 by 𝑤 = (𝑢, −𝑣). Show
that the map 𝑤 ↦ 𝑤 is a bijective conjugate-linear map from 𝑉ℂ to itself
satisfying 𝑤 = 𝑤 for all 𝑤 ∈ 𝑉ℂ . (A map 𝐹 ∶ 𝑉 → 𝑊 between complex
vector spaces is said to be conjugate-linear if it is linear over ℝ and satisfies
𝐹 (𝛼𝑣) = 𝛼𝐹 (𝑣) for all 𝛼 ∈ ℂ and 𝑣 ∈ 𝑉 .)
(b) An element 𝑤 ∈ 𝑉ℂ is said to be real if 𝑤 = 𝑤. Show that 𝑤 is real if and
only if it lies in the real subspace 𝑉 ⊆ 𝑉ℂ defined above.
(c) For 𝑤 ∈ 𝑉ℂ , define Re 𝑤 = 21 (𝑤 + 𝑤) and Im 𝑤 = 2𝑖1 (𝑤 − 𝑤). Show that
Re 𝑤 and Im 𝑤 are real, and 𝑤 = Re 𝑤 + 𝑖 Im 𝑤.

The complexification functor can be adapted easily to vector bundles. First we

Suppose 𝑀 is a topological space. A complex vector bundle of rank 𝒌 over

establish some definitions.

𝑴 is defined analogously to a real vector bundle (e.g., as in [LeeSM, Chap. 10]):

it is a topological space 𝐸 together with a continuous surjective map 𝜋 ∶ 𝐸 → 𝑀
such that each fiber 𝐸𝑝 = 𝜋 −1 (𝑝) is given the structure of a 𝑘-dimensional complex
vector space, and each 𝑝 ∈ 𝑀 has a neighborhood 𝑈 over which there exists a local
trivialization, which is a homeomorphism Φ ∶ 𝜋 −1 (𝑈 ) → 𝑈 × ℂ𝑘 that restricts to
a complex-linear isomorphism from 𝐸𝑞 to {𝑞} × ℂ𝑘 for each 𝑞 ∈ 𝑈 . This means,
in particular, that the following diagram commutes, where 𝜋1 ∶ 𝑈 × ℂ𝑘 → 𝑈 is the
projection on the first factor:

Φ
𝜋 −1 (𝑈●) / 𝑈 × ℂ𝑘

𝜋|𝜋 −1 (𝑈 ) ●●# ✇✇ 𝜋1
●● ✇✇
●● ✇✇

𝑈.
●● ✇
{✇✇

If 𝑀 and 𝐸 are smooth manifolds, 𝜋 is a smooth map, and the local trivializations

if 𝑀 and 𝐸 are complex manifolds, 𝜋 is holomorphic, and the local trivializations

can be chosen to be diffeomorphisms, it is a smooth complex vector bundle; and
The Complexified Tangent and Cotangent Bundles 25

open cover of 𝑀 such that 𝐸 admits a trivialization over each of the open sets of
can be chosen to be biholomorphisms, it is a holomorphic vector bundle. Any

the cover is called a trivializing cover for 𝑬. If there is a global trivialization (that
is, a local trivialization over all of 𝑀), the bundle is said to be a trivial bundle. A
line bundle is a (real or complex) vector bundle of rank 1.
If 𝜋 ∶ 𝐸 → 𝑀 and 𝜋 ′ ∶ 𝐸 ′ → 𝑀 are complex vector bundles over 𝑀, a map
𝐹 ∶ 𝐸 → 𝐸 ′ is called a bundle homomorphism if 𝜋 ′ ∘ 𝐹 = 𝜋 and for each 𝑝 ∈ 𝑀,
the map 𝐹 |𝐸𝑝 ∶ 𝐸𝑝 → 𝐸𝑝′ is a complex-linear map. A bundle homomorphism that is
also a homeomorphism between 𝐸 and 𝐸 ′ is called a bundle isomorphism, and the
bundles 𝐸 and 𝐸 ′ are said to be isomorphic, denoted by 𝐸 ≅ 𝐸 ′ , if there is a bundle
isomorphism between them. If the bundles are smooth and 𝐹 is a diffeomorphism,
it is called a smooth isomorphism, and if the bundles are holomorphic and 𝐹 is a
biholomorphism, it is a holomorphic isomorphism. In each of these cases, it is easy
to check that the inverse map is also a bundle isomorphism. (For some purposes,
it is useful to introduce a more general notion of vector bundle homomorphisms

identified as a bundle homomorphism over 𝑴; see [LeeSM, Chap. 10] for details.
between bundles over different manifolds, and the kind we have defined here is

Since we will not have any need for that extra generality, we always understand
bundle homomorphisms to be the type we have defined here.)
Most of the standard constructions used for real vector bundles, such as Whit-
ney sums [LeeSM, Example 10.7] and smooth subbundles [LeeSM, pp. 264–266],
carry over in obvious ways to smooth complex bundles.
We will have much more to say about holomorphic vector bundles in Chapter

If 𝜋 ∶ 𝐸 → 𝑀 is a smooth (real or complex) vector bundle, a (global) section

3; for now we focus attention on smooth bundles.

of 𝑬 is a continuous map 𝜎 ∶ 𝑀 → 𝐸 such that 𝜋 ∘ 𝜎 = Id𝑀 . For any open subset

𝑈 ⊆ 𝑀, a local section of 𝑬 over 𝑼 is a continuous map 𝜎 ∶ 𝑈 → 𝐸 satisfying
𝜎 ∘ 𝜋 = Id𝑈 . Every smooth vector bundle has a smooth zero section 𝜁, for which
𝜁(𝑝) is the zero element of 𝐸𝑝 for each 𝑝 ∈ 𝑀. Any section that is not equal to the

𝑬 is a map 𝜎 ∶ 𝑈 → 𝐸 rough (local or global) section of 𝑬 is a map 𝜎 ∶ 𝑈 → 𝐸

zero section will be called a nontrivial section. A rough (local or global) section of

satisfying 𝜎 ∘ 𝜋 = Id𝑈 , but not assumed to be smooth or even continuous. We

denote the space of smooth global sections of 𝐸 by Γ(𝐸). A local frame for 𝑬 is
an ordered 𝑘-tuple of local sections (𝜎1 , … , 𝜎𝑘 ) over an open set 𝑈 ⊆ 𝑀 whose
values at each 𝑝 ∈ 𝑈 form a basis for the fiber 𝐸𝑝 .
If 𝜋 ∶ 𝐸 → 𝑀 is a smooth rank-𝑘 real vector bundle over a smooth manifold
𝑀, we define the complexification of 𝑬 to be the set 𝐸ℂ = ⋃𝑝∈𝑀 (𝐸𝑝 )ℂ together
with the obvious projection 𝜋ℂ ∶ 𝐸ℂ → 𝑀. For each smooth local trivialization
Φ ∶ 𝜋 −1 (𝑈 ) → 𝑈 × ℝ𝑘 , we define a local trivialization Φℂ ∶ 𝜋ℂ−1 (𝑈 ) → 𝑈 × ℂ𝑘 by

Φℂ (𝜉) = (𝜋ℂ (𝜉), (Φ|𝐸𝜋 (𝜉) )ℂ (𝜉)).

ℂ
26 1. The Basics

Wherever two such trivializations (𝑈 , Φ) and (𝑉 , Ψ) overlap, [LeeSM, Lemma

10.15] shows that we can write Ψ ∘ Φ−1 (𝑝, 𝑣) = (𝑝, 𝜏(𝑝)𝑣) for some smooth tran-
sition function 𝜏 ∶ 𝑈 ∩ 𝑉 → GL(𝑘, ℝ), and it is straightforward to check that the
transition function from Φℂ to Ψℂ is the same: Ψℂ ∘ Φ−1ℂ (𝑝, 𝑣) = (𝑝, 𝜏(𝑝)𝑣), where
now we are considering 𝜏 as a map into GL(𝑘, ℂ). It follows from the vector bundle

tor bundles) that 𝜋ℂ ∶ 𝐸ℂ → 𝑀 has a unique structure as a smooth rank-𝑘 complex

chart lemma [LeeSM, Lemma 10.6] (adapted in the obvious way for complex vec-

vector bundle, with the maps constructed above as smooth local trivializations.

(𝑏1 , … , 𝑏𝑘 ) for 𝐸, we can write a section of 𝐸ℂ locally as a sum 𝑓 𝑗 𝑏𝑗 , where now

What this really amounts to in practice is that, given any smooth local frame

the coefficient functions 𝑓 𝑗 are allowed to be complex-valued.

► Exercise 1.40. Let 𝐸 → 𝑀 be a smooth real vector bundle. Show that every
smooth (local or global) section of 𝐸ℂ can be written uniquely as a sum 𝛼 + 𝑖𝛽,
where 𝛼 and 𝛽 are smooth local or global sections of 𝐸.

The result of Exercise 1.39 shows that for any real vector bundle 𝐸 → 𝑀,
conjugation defines a smooth conjugate-linear bundle homomorphism from 𝐸ℂ to
itself, and the set of real elements (those satisfying 𝑤 = 𝑤) forms a real-linear
subbundle canonically isomorphic to the original bundle 𝐸. It is important to note
that the existence of such a conjugation operator is a special feature of complex-
ifications: in fact, as Problem 1-6 shows, a complex vector bundle admits such a
conjugation operator if and only if it is isomorphic to the complexification of a real
bundle.

smooth manifold 𝑀, we obtain the complexified tangent bundle 𝑇ℂ 𝑀 and the

When we apply this construction to the tangent and cotangent bundles of a

complexified cotangent bundle 𝑇ℂ∗ 𝑀, respectively. A section of 𝑇ℂ 𝑀, called a

complex vector field, can be written locally as a linear combination of coordinate

field plus 𝑖 times another real vector field. A complex vector field 𝑍 = 𝑋 + 𝑖𝑌
vector fields with complex-valued coefficient functions, or as a sum of a real vector

acts on a smooth real-valued function 𝑓 by 𝑍𝑓 = 𝑋𝑓 + 𝑖𝑌 𝑓 , and on a complex-

valued function 𝑓 = 𝑢 + 𝑖𝑣 by the same formula, where we interpret 𝑋𝑓 to mean
𝑋𝑢 + 𝑖𝑋𝑣 and similarly for 𝑌 . The Lie bracket operation can be extended to pairs
of smooth complex vector fields by complex bilinearity: [𝑋1 + 𝑖𝑌1 , 𝑋2 + 𝑖𝑌2 ] =
([𝑋1 , 𝑋2 ] − [𝑌1 , 𝑌2 ]) + 𝑖([𝑋1 , 𝑌2 ] + [𝑌2 , 𝑋1 ]). It is straightforward to check that
the formula [𝑓 𝑉 , 𝑔𝑊 ] = 𝑓 𝑔[𝑉 , 𝑊 ] + 𝑓 (𝑉 𝑔)𝑊 − 𝑔(𝑊 𝑓 )𝑉 holds equally well
when the vector fields 𝑉 , 𝑊 and the functions 𝑓 , 𝑔 are allowed to be complex.
Similarly, a section of 𝑇ℂ∗ 𝑀 is called a complex 1-form or a complex covector
field, and can be written locally as a linear combination of coordinate 1-forms with
complex coefficients, or as a sum of a real 1-form plus 𝑖 times another real 1-form.
With this construction, we are now justified in writing 𝑑𝑓 = 𝑑𝑢 + 𝑖 𝑑𝑣 whenever
𝑓 = 𝑢 + 𝑖𝑣 is a complex-valued smooth function.
The Complexified Tangent and Cotangent Bundles 27

► Exercise 1.41. Prove that there is a canonical smooth bundle isomorphism

between 𝑇ℂ∗ 𝑀 and the bundle Homℂ (𝑇ℂ 𝑀, ℂ) whose fiber at a point 𝑝 ∈ 𝑀 is
the space of complex-linear maps from (𝑇𝑝 𝑀)ℂ to ℂ.

Let us specialize to the case of ℂ𝑛 , with its standard holomorphic coordinates

𝑧 = 𝑥𝑗 + 𝑖𝑦𝑗 . Considering ℂ𝑛 as a smooth manifold of (real) dimension 2𝑛, we
𝑗

can use (𝑥𝑗 , 𝑦𝑗 ) as smooth global coordinates. We have a smooth global coframe
{𝑑𝑥𝑗 , 𝑑𝑦𝑗 } for 𝑇 ∗ ℂ𝑛 , which is therefore also a coframe for 𝑇ℂ∗ ℂ𝑛 . Consider the 2𝑛
complex 1-forms 𝑑𝑧𝑗 = 𝑑𝑥𝑗 + 𝑖 𝑑𝑦𝑗 and 𝑑𝑧𝑗 = 𝑑𝑥𝑗 − 𝑖 𝑑𝑦𝑗 . Because we can solve
for 𝑑𝑥𝑗 = 21 (𝑑𝑧𝑗 + 𝑑𝑧𝑗 ) and 𝑑𝑦𝑗 = 2𝑖1 (𝑑𝑧𝑗 − 𝑑𝑧𝑗 ), it follows that {𝑑𝑧𝑗 , 𝑑𝑧𝑗 } is also
a smooth coframe for 𝑇ℂ∗ ℂ𝑛 , and arbitrary complex 1-forms can also be expressed
in terms of this coframe. In particular, if 𝑓 ∶ 𝑈 → ℂ is a smooth function on an
open subset 𝑈 ⊆ ℂ𝑛 , we can write
𝜕𝑓 𝜕𝑓
𝑑𝑓 = 𝑑𝑥𝑗 + 𝑗 𝑑𝑦𝑗 = 𝐴𝑗 𝑑𝑧𝑗 + 𝐵𝑗 𝑑𝑧𝑗
𝜕𝑥 𝑗 𝜕𝑦
for some coefficient functions 𝐴𝑗 and 𝐵𝑗 . (When using the summation convention,

lower index.) To see what these coefficients are, just substitute the formulas for 𝑑𝑥𝑗
the understanding is that an upper index “in the denominator” is to be treated as a

and 𝑑𝑦𝑗 in terms of 𝑑𝑧𝑗 , 𝑑𝑧𝑗 and collect terms:

𝜕𝑓 𝑑𝑧𝑗 + 𝑑𝑧𝑗 𝜕𝑓 𝑑𝑧𝑗 − 𝑑𝑧𝑗
𝑑𝑓 = +
𝜕𝑥𝑗 ( 2 ) 𝜕𝑥𝑗 ( 2𝑖 )
1 𝜕𝑓 𝜕𝑓 1 𝜕𝑓 𝜕𝑓
= − 𝑖 𝑗 𝑑𝑧𝑗 + + 𝑖 𝑗 𝑑𝑧𝑗 .
(1.7)

2 ( 𝜕𝑥𝑗 𝜕𝑦 ) 2 ( 𝜕𝑥 𝑗 𝜕𝑦 )

Motivated by this calculation, we define 2𝑛 smooth complex vector fields 𝜕/𝜕𝑧𝑗

and 𝜕/𝜕𝑧𝑗 on ℂ𝑛 by
𝜕 1 𝜕 𝜕 𝜕 1 𝜕 𝜕
= −𝑖 𝑗 , = +𝑖 𝑗 .
𝜕𝑧 𝑗 (
2 𝜕𝑥 𝑗 𝜕𝑦 ) 𝜕𝑧 𝑗 (
2 𝜕𝑥 𝑗 𝜕𝑦 )
(1.8)

(Be sure to notice that the negative sign appears in the formula for 𝜕/𝜕𝑧𝑗 , not 𝜕/𝜕𝑧𝑗 ;
this is not a typo!) A simple computation shows that {𝜕/𝜕𝑧𝑗 , 𝜕/𝜕𝑧𝑗 } is the smooth
global frame for 𝑇ℂ ℂ𝑛 dual to {𝑑𝑧𝑗 , 𝑑𝑧𝑗 }. For a smooth complex-valued function
𝑓 defined on an open subset 𝑈 ⊆ ℂ𝑛 , formula (1.7) can be rewritten in terms of

𝜕𝑓 𝜕𝑓
this frame as
𝑑𝑓 = 𝑗 𝑑𝑧𝑗 + 𝑗 𝑑𝑧𝑗 .
𝜕𝑧 𝜕𝑧
(1.9)

In the special case in which 𝑓 is a holomorphic function on an open subset of

ℂ𝑛 , you will notice that we had already defined the expression 𝜕𝑓 /𝜕𝑧𝑗 by equation
(1.2); now we seem to have introduced a different meaning for the same expression.
The next proposition ensures that the two definitions are equivalent for holomorphic
functions.
28 1. The Basics

Proposition 1.42. Suppose 𝑈 ⊆ ℂ𝑛 is open. Let 𝑓 ∶ 𝑈 → ℂ be any smooth

function, and let 𝜕/𝜕𝑧𝑗 and 𝜕/𝜕𝑧𝑗 be the complex vector fields on 𝑈 defined by (1.8).
(a) 𝑓 is holomorphic if and only if 𝜕𝑓 /𝜕𝑧𝑗 = 0 for 𝑗 = 1, … , 𝑛.
(b) If 𝑓 is holomorphic, then for each 𝑗, the expression 𝜕𝑓 /𝜕𝑧𝑗 obtained by
applying the complex vector field 𝜕/𝜕𝑧𝑗 to 𝑓 is equal to the complex partial
derivative defined by (1.2).

Proof. After we substitute 𝑓 = 𝑢 + 𝑖𝑣 into the equation 𝜕𝑓 /𝜕𝑧𝑗 = 0 and separate

its real and imaginary parts, it becomes the 𝑗th pair of Cauchy–Riemann equations
for 𝑓 , thus proving (a). Then (b) follows from Proposition 1.25. □

One must be careful not to read too much into the expressions 𝜕𝑓 /𝜕𝑧𝑗 and
𝜕𝑓 /𝜕𝑧𝑗 when 𝑓 is merely smooth: despite the notation, they are not partial deriva-

derivative of a function with respect to 𝑧1 while holding 𝑧2 , … , 𝑧𝑛 , 𝑧1 , … , 𝑧𝑛 fixed.

tives in the ordinary sense, because, for example, it does not make sense to take a

If you fix 𝑧1 , then 𝑧1 remains fixed as well. However, there is a sense in which these

Suppose 𝑝 is any (not necessarily holomorphic) complex-valued polynomial

operators behave like partial derivatives, which we now explain.

function of the real variables {𝑥𝑗 , 𝑦𝑗 }:

𝑝(𝑥, 𝑦) = 𝑎𝑙1 ,…,𝑙𝑛 ,𝑚1 ,…,𝑚𝑛 (𝑥1 )𝑙1 ⋯ (𝑥𝑛 )𝑙𝑛 (𝑦1 )𝑚1 ⋯ (𝑦𝑛 )𝑚𝑛 .
∑
𝑙1 ,…,𝑙𝑛
𝑚1 ,…,𝑚𝑛

Substituting 𝑥𝑗 = 21 (𝑧𝑗 + 𝑧𝑗 ) and 𝑦𝑗 = 2𝑖1 (𝑧𝑗 − 𝑧𝑗 ) and collecting like terms, we can
express 𝑝 as a polynomial expression in 𝑧𝑗 , 𝑧𝑗 , which we denote by 𝑝:̃
𝑧+𝑧 𝑧−𝑧
𝑝(𝑧)
̃ = 𝑝( ,
2 2𝑖 )
= 𝑎̃ (𝑧1 )𝑙1 ⋯ (𝑧𝑛 )𝑙𝑛 (𝑧1 )𝑚1 ⋯ (𝑧𝑛 )𝑚𝑛 .
∑ 𝑙1 ,…,𝑙𝑛 ,𝑚1 ,…,𝑚𝑛
𝑙1 ,…,𝑙𝑛
𝑚1 ,…,𝑚𝑛

To separate the dependence on 𝑧𝑗 and 𝑧𝑗 , we can introduce new independent

variables 𝑤𝑗 in place of 𝑧𝑗 . Let 𝑞 ∶ ℂ𝑛 × ℂ𝑛 → ℂ be the polynomial function
𝑧+𝑤 𝑧−𝑤
𝑞(𝑧, 𝑤) = 𝑝 ( , ,
2 2𝑖 )
so that 𝑝(𝑧)
̃ = 𝑞(𝑧, 𝑧). Now it makes sense to ask whether 𝑞 is independent of
𝑤 1 , … , 𝑤𝑛 .

► Exercise 1.43. Prove that the original polynomial 𝑝 defines a holomorphic

function if and only if 𝜕𝑞/𝜕𝑤𝑗 = 0 for each 𝑗.

So for a polynomial function 𝑝, in this sense we can say 𝑝 is holomorphic if and

only if it depends only on 𝑧1 , … , 𝑧𝑛 with no occurrences of 𝑧1 , … , 𝑧𝑛 . Exactly the
The Complexified Tangent and Cotangent Bundles 29

same argument can be made when 𝑝 is a real-analytic function, except then the finite
sums above become absolutely convergent infinite series; the absolute convergence

as well, a real-analytic function 𝑓 is holomorphic if and only if it can be written as

ensures that the convergence is not affected by rearranging the terms. In that case

a power series in 𝑧1 , … , 𝑧𝑛 , with no occurrences of 𝑧1 , … , 𝑧𝑛 .

For a function that is merely smooth, these computations do not make sense,

real values of (𝑥1 , … , 𝑥𝑛 , 𝑦1 , … , 𝑦𝑛 ). But motivated by the computations above, it

because you cannot plug complex numbers into a function that is defined only for

function that is independent of 𝑧1 , … , 𝑧𝑛 .”

is sometimes helpful to think about a holomorphic function intuitively as a “smooth

Complex Coordinate Frames

Now suppose 𝑀 is a complex manifold and (𝑧1 , … , 𝑧𝑛 ) are local holomor-
phic coordinates on an open subset 𝑈 ⊆ 𝑀. The coordinate map 𝜑 ∶ 𝑈 → ℂ𝑛
can also be thought of as a smooth coordinate map from 𝑈 to ℝ2𝑛 , with smooth
coordinate functions (𝑥1 , 𝑦1 , … , 𝑥𝑛 , 𝑦𝑛 ) where 𝑧𝑗 = 𝑥𝑗 + 𝑖𝑦𝑗 . These coordinates
yield smooth coordinate vector fields (𝜕/𝜕𝑥1 , 𝜕/𝜕𝑦1 , … , 𝜕/𝜕𝑥𝑛 , 𝜕/𝜕𝑦𝑛 ), which act on
a smooth function 𝑓 ∶ 𝑈 → ℂ by
𝜕 𝜕 𝜕 𝜕
𝑓= (𝑓 ∘ 𝜑−1 ), 𝑓= (𝑓 ∘ 𝜑−1 ),
𝜕𝑥𝑗 |𝑝 𝜕𝑥𝑗 |𝜑(𝑝) 𝜕𝑦𝑗 |𝑝 𝜕𝑦𝑗 |𝜑(𝑝)
(1.10)

ℝ2𝑛 (see [LeeSM, p. 60]). We define a smooth local complex frame {𝜕/𝜕𝑧𝑗 , 𝜕/𝜕𝑧𝑗 }
where the expressions on the right-hand sides are ordinary partial derivatives on

for 𝑇ℂ 𝑀 by (1.8), where now 𝜕/𝜕𝑥𝑗 and 𝜕/𝜕𝑦𝑗 are interpreted as smooth vector
fields on 𝑈 ⊆ 𝑀. These vector fields are called complex coordinate vector fields,
and the corresponding local frame is called a complex coordinate frame.
Lemma 1.44. Suppose 𝑀 is a complex manifold and 𝑓 ∶ 𝑀 → ℂ is a smooth
function. If (𝑧1 , … , 𝑧𝑛 ) are holomorphic coordinates on a subset 𝑈 ⊆ 𝑀 and
{𝜕/𝜕𝑧𝑗 , 𝜕/𝜕𝑧𝑗 } are the corresponding complex coordinate vector fields, then 𝑓 is
holomorphic on 𝑈 if and only if 𝜕𝑓 /𝜕𝑧𝑗 ≡ 0 on 𝑈 for 𝑗 = 1, … , 𝑛.

Proof. Let 𝜑 ∶ 𝑈 → ℂ𝑛 be the holomorphic coordinate map, and let 𝑈 ̂ = 𝜑(𝑈 ) ⊆

ℂ . It follows from (1.10) together with the definition of 𝜕/𝜕𝑧 that for all 𝑝 ∈ 𝑈 ,
𝑛 𝑗

𝜕𝑓 𝜕(𝑓 ∘ 𝜑−1 )
(𝑝) = (𝜑(𝑝)).
𝜕𝑧𝑗 𝜕𝑧𝑗
The lemma then follows from the fact that 𝑓 ∶ 𝑈 → ℂ is holomorphic by definition
̂ → ℂ is holomorphic.
if and only if 𝑓 ∘ 𝜑−1 ∶ 𝑈 □

When 𝑀 and 𝑁 are complex manifolds, the total derivative or differential

of a smooth map 𝐹 ∶ 𝑀 → 𝑁 at a point 𝑝 ∈ 𝑀 is a real-linear map from
𝑇𝑝 𝑀 to 𝑇𝐹 (𝑝) 𝑁, and its complexification is a complex-linear map from (𝑇𝑝 𝑀)ℂ
30 1. The Basics

to (𝑇𝐹 (𝑝) 𝑁)ℂ . For smooth manifolds, the differential is often denoted by 𝑑𝐹𝑝 , but

at 𝑝 (or its complexification) by 𝐷𝐹 (𝑝), and the associated bundle homomorphism,

for reasons that will be explained shortly, in this book we will denote the differential

called the global differential of 𝑭 , by 𝐷𝐹 ∶ 𝑇ℂ 𝑀 → 𝑇ℂ 𝑁. The next proposition

shows how to compute it in terms of holomorphic coordinates.
Proposition 1.45 (The Total Derivative in Holomorphic Coordinates). Let 𝑀
and 𝑁 be complex manifolds and 𝐹 ∶ 𝑀 → 𝑁 be a smooth map. Given 𝑝 ∈ 𝑀,
let 𝑧𝑗 = 𝑥𝑗 + 𝑖𝑦𝑗 be local holomorphic coordinates for 𝑀 in a neighborhood of 𝑝,
and 𝑤𝑗 = 𝑢𝑗 + 𝑖𝑣𝑗 for 𝑁 in a neighborhood of 𝐹 (𝑝). In terms of the complex local
frames {𝜕/𝜕𝑧𝑗 , 𝜕/𝜕𝑧𝑗 } for 𝑀 and {𝜕/𝜕𝑤𝑗 , 𝜕/𝜕𝑤𝑗 } for 𝑁, the total derivative of 𝐹
at 𝑝 has the following coordinate representation:

𝜕 𝜕𝐹 𝑘 𝜕 𝜕𝐹 𝑘 𝜕
𝐷𝐹 (𝑝)( | ) = (𝑝) + (𝑝) ,
𝜕𝑧 𝑝
𝑗 𝜕𝑧𝑗 𝜕𝑤𝑘 |𝐹 (𝑝) 𝜕𝑧𝑗 𝜕𝑤𝑘 |𝐹 (𝑝)
(1.11)

𝜕 𝜕𝐹 𝑘 𝜕 𝜕𝐹 𝑘 𝜕
𝐷𝐹 (𝑝)( 𝑗 | ) = (𝑝) + (𝑝) .
𝜕𝑧 𝑝 𝜕𝑧𝑗 𝜕𝑤𝑘 |𝐹 (𝑝) 𝜕𝑧𝑗 𝜕𝑤𝑘 |𝐹 (𝑝)
(1.12)

Proof. Write the real and imaginary parts of (the coordinate representation of) 𝐹
as 𝐹 = 𝑈 + 𝑖𝑉 . Considering 𝑀 and 𝑁 as smooth manifolds, we have the usual
coordinate formula for 𝐷𝐹 (𝑝):

𝜕 𝜕𝑈 𝑘 𝜕 𝜕𝑉 𝑘 𝜕
𝐷𝐹 (𝑝)( = (𝑝) 𝑘 | + (𝑝) 𝑘 | ,
𝜕𝑥𝑗 |𝑝 ) 𝜕𝑥𝑗 𝜕𝑢 𝐹 (𝑝) 𝜕𝑥𝑗 𝜕𝑣 𝐹 (𝑝)
𝜕 𝜕𝑈 𝑘 𝜕 𝜕𝑉 𝑘 𝜕
𝐷𝐹 (𝑝)( 𝑗 ) = (𝑝) 𝑘 | + (𝑝) 𝑘 | .
(1.13)

𝜕𝑦 |𝑝 𝜕𝑦𝑗 𝜕𝑢 𝐹 (𝑝) 𝜕𝑦𝑗 𝜕𝑣 𝐹 (𝑝)

To transform this to holomorphic coordinates, begin with the definitions of 𝜕/𝜕𝑧𝑗

and 𝜕/𝜕𝑧𝑗 , and use (1.13) together with the complex linearity of 𝐷𝐹 (𝑝) to obtain

𝜕 𝜕𝑈 𝑘 𝜕 𝜕𝑉 𝑘 𝜕
𝐷𝐹 (𝑝)( | ) = (𝑝) 𝑘 | + (𝑝) 𝑘 | ,
𝜕𝑧 𝑝
𝑗 𝜕𝑧 𝑗 𝜕𝑢 𝐹 (𝑝) 𝜕𝑧𝑗 𝜕𝑣 𝐹 (𝑝)
𝜕 𝜕𝑈 𝑘 𝜕 𝜕𝑉 𝑘 𝜕
𝐷𝐹 (𝑝)( 𝑗 | ) = (𝑝) 𝑘 | + (𝑝) 𝑘 | .
𝜕𝑧 𝑝 𝜕𝑧𝑗 𝜕𝑢 𝐹 (𝑝) 𝜕𝑧𝑗 𝜕𝑣 𝐹 (𝑝)

Now substitute 𝜕/𝜕𝑢𝑘 = 𝜕/𝜕𝑤𝑘 + 𝜕/𝜕𝑤𝑘 and 𝜕/𝜕𝑣𝑘 = 𝑖(𝜕/𝜕𝑤𝑘 − 𝜕/𝜕𝑤𝑘 ) and collect
terms:

𝜕 𝜕𝑈 𝑘 𝜕𝑉 𝑘 𝜕
𝐷𝐹 (𝑝)( | ) = ( (𝑝) + 𝑖 (𝑝))
𝜕𝑧 𝑝
𝑗 𝜕𝑧 𝑗 𝜕𝑧 𝑗 𝜕𝑤𝑘 |𝐹 (𝑝)
𝜕𝑈 𝑘 𝜕𝑉 𝑘 𝜕
+ ( 𝑗 (𝑝) − 𝑖 𝑗 (𝑝)) .
𝜕𝑧 𝜕𝑧 𝜕𝑤𝑘 |𝐹 (𝑝)
This is (1.11), and a similar computation proves (1.12). □
The Complexified Tangent and Cotangent Bundles 31

Corollary 1.46. In addition to the hypotheses of 1.45, suppose 𝐹 is holomorphic.

Then in terms of the local frames {𝜕/𝜕𝑧𝑗 , 𝜕/𝜕𝑧𝑗 } and {𝜕/𝜕𝑤𝑗 , 𝜕/𝜕𝑤𝑗 }, 𝐷𝐹 (𝑝) is
represented by the block-diagonal matrix
𝐷′ 𝐹 (𝑝) 0
,
( 0 𝐷 𝐹 (𝑝))
′
(1.14)

where 𝐷′ 𝐹 denotes the 𝑛 × 𝑛 complex matrix-valued function (𝜕𝐹 𝑘 /𝜕𝑧𝑗 ), called

the holomorphic Jacobian of 𝑭 . Thus the linear map 𝐷𝐹 (𝑝) is invertible if and
only if the holomorphic Jacobian of 𝐹 is invertible at 𝑝.

Proof. The fact that 𝐹 is holomorphic means that each component function of its
coordinate representation is holomorphic. Thus 𝜕𝐹 𝑘 /𝜕𝑧𝑗 vanishes identically, and
by conjugation so does 𝜕𝐹 𝑘 /𝜕𝑧𝑗 . Therefore, 𝐷𝐹 (𝑝) has the given matrix repre-

det 𝐷𝐹 (𝑝) = | det 𝐷′ 𝐹 (𝑝)|2 .

sentation by Proposition 1.45. The last statement then follows from the fact that

Proposition 1.47 (Chain Rule for Smooth Functions). Suppose 𝑀 and 𝑁 are
□

complex manifolds, 𝐹 ∶ 𝑀 → 𝑁 is a smooth map, and ℎ ∶ 𝑁 → ℂ is a smooth

function. In terms of local holomorphic coordinates (𝑧𝑗 ) for 𝑀 and (𝜁 𝑘 ) for 𝑁,
𝜕(ℎ ∘ 𝐹 ) 𝜕ℎ 𝜕𝐹 𝑘 𝜕ℎ 𝜕𝐹 𝑘
= + 𝑘 ,
𝜕𝑧𝑗 𝜕𝜁 𝑘 𝜕𝑧𝑗 𝜕𝜁 𝜕𝑧𝑗
𝜕(ℎ ∘ 𝐹 ) 𝜕ℎ 𝜕𝐹 𝑘 𝜕ℎ 𝜕𝐹 𝑘
= + 𝑘 .
𝜕𝑧 𝑗 𝜕𝜁 𝜕𝑧
𝑘 𝑗 𝜕𝜁 𝜕𝑧𝑗

Proof. Proposition 1.45 shows that the value of 𝜕(ℎ ∘ 𝐹 )/𝜕𝑧𝑗 at 𝑝 ∈ 𝑀 is equal to
the 𝜕/𝜕𝑤 component of 𝐷(ℎ ∘ 𝐹 )(𝑝)(𝜕/𝜕𝑧𝑗 |𝑝 ) (where 𝑤 denotes the standard holo-
morphic coordinate of ℂ). By smooth manifold theory, 𝐷(ℎ ∘ 𝐹 )(𝑝) = 𝐷ℎ(𝐹 (𝑝)) ∘
𝐷𝐹 (𝑝), which can be computed by applying the formula of Proposition 1.45 to ℎ
and to 𝐹 and composing the two linear maps. A similar argument applies to the 𝑧𝑗
derivative. □

of Proposition 1.47, suppose in addition that 𝐹 and ℎ are holomorphic. Then

Corollary 1.48 (Chain Rule for Holomorphic Functions). Under the hypotheses

𝜕ℎ 𝜕𝐹 𝑘 𝑗
𝑑(ℎ ∘ 𝐹 ) = 𝑑𝑧 .
𝜕𝑤𝑘 𝜕𝑧𝑗
□

function 𝑓 at a point 𝑝 ∈ 𝑀 can be considered either as a linear map from 𝑇𝑝 𝑀

In the theory of smooth (real) manifolds, the differential of a smooth real-valued

to ℝ (a covector) or as a linear map from 𝑇𝑝 𝑀 to 𝑇𝑓 (𝑝) ℝ; in view of the canonical

identification between 𝑇𝑓 (𝑝) ℝ and ℝ, these are the same map, so it makes sense to
use the same notation 𝑑𝑓𝑝 to denote both of them. But in complex manifold theory,
something different happens. Suppose 𝑓 = 𝑢 + 𝑖𝑣 ∶ 𝑀 → ℂ is a complex-valued
smooth function on a complex manifold 𝑀. On the one hand, 𝑑𝑓𝑝 denotes the value
at 𝑝 of the complex-valued 1-form 𝑑𝑓 = 𝑑𝑢 + 𝑖𝑑𝑣, an element of (𝑇𝑝∗ 𝑀)ℂ , which
32 1. The Basics

can also be viewed as a complex-linear map from (𝑇𝑝 𝑀)ℂ to ℂ (by Exercise 1.41).
Using the coordinate formula (1.9), we find, for example, that
𝜕 𝜕𝑓
𝑑𝑓𝑝 ( | ) = 𝑗 (𝑝) ∈ ℂ.
𝜕𝑧 𝑝
𝑗 𝜕𝑧
On the other hand, 𝐷𝑓 (𝑝) is a complex-linear map from (𝑇𝑝 𝑀)ℂ to (𝑇𝑓 (𝑝) ℂ)ℂ , and
Proposition 1.45 shows that
𝜕 𝜕𝑓 𝜕 𝜕𝑓 𝜕
𝐷𝑓 (𝑝)( | ) = 𝑗 (𝑝) | + 𝑗 (𝑝) |𝑓 (𝑝) ∈ (𝑇𝑓 (𝑝) ℂ)ℂ .
𝜕𝑧 𝑝
𝑗 𝜕𝑧 𝜕𝑤 𝑓 (𝑝) 𝜕𝑧 𝜕𝑤
These are distinctly different objects—for example, if 𝑓 is holomorphic, then
𝑑𝑓 (𝜕/𝜕𝑧𝑗 ) vanishes identically, but 𝐷𝑓 (𝜕/𝜕𝑧𝑗 ) does not. This is why we use differ-

for 𝐷𝑓 (𝑝).
ent notations for the two kinds of derivatives, and prefer the term “total derivative”

Orientations
The computations we just did lead to another important property of complex
manifolds: they all have canonical orientations. (Just to be clear: when we speak
of an orientation of a complex manifold, it means an orientation of its underlying
smooth real manifold.)
Proposition 1.49. Every complex manifold has a canonical orientation, uniquely
determined by the following two properties:
(i) The canonical orientation of ℂ𝑛 is the one determined by the 2𝑛-form

(1.15) 𝜔𝑛 = 𝑑𝑥1 ∧ 𝑑𝑦1 ∧ ⋯ ∧ 𝑑𝑥𝑛 ∧ 𝑑𝑦𝑛 .

(ii) Every local biholomorphism is orientation-preserving.

Proof. Let us begin by expressing the real 2𝑛-form 𝜔𝑛 in terms of the complex co-
ordinates (𝑧1 , … , 𝑧𝑛 ). Observe that for each 𝑗, we have 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑗 = (𝑑𝑥𝑗 + 𝑖 𝑑𝑦𝑗 ) ∧
(𝑑𝑥𝑗 − 𝑖 𝑑𝑦𝑗 ) = −2𝑖𝑑𝑥𝑗 ∧ 𝑑𝑦𝑗 . Therefore
𝑖 𝑛
𝜔𝑛 = ( ) 𝑑𝑧1 ∧ 𝑑𝑧1 ∧ ⋯ ∧ 𝑑𝑧𝑛 ∧ 𝑑𝑧𝑛 .
2
(1.16)

Let 𝑈 ⊆ ℂ𝑛 be an open subset and 𝐹 ∶ 𝑈 → ℂ𝑛 be a local biholomorphism.

The Jacobian matrix of 𝐹 has the form (1.14) when expressed in terms of the or-
dered frame (𝜕/𝜕𝑧1 , … , 𝜕/𝜕𝑧𝑛 , 𝜕/𝜕𝑧1 , … , 𝜕/𝜕𝑧𝑛 ). Note that this order is not the

is (−1)(𝑛−1)𝑛/2 , as you can check, and therefore

same as the one used in formula (1.16)—they differ by a permutation whose sign

𝑖 𝑛
𝜔𝑛 = (−1)(𝑛−1)𝑛/2 ( ) 𝑑𝑧1 ∧ ⋯ ∧ 𝑑𝑧𝑛 ∧ 𝑑𝑧1 ∧ ⋯ ∧ 𝑑𝑧𝑛 .
2
The Complexified Tangent and Cotangent Bundles 33

The formula for the pullback of a top-degree form (see [LeeSM, Prop. 14.9],
which works equally well for complex-valued forms) gives

𝐹 ∗ (𝑑𝑧1 ∧ ⋯ ∧ 𝑑𝑧𝑛 ∧ 𝑑𝑧1 ∧ ⋯ ∧ 𝑑𝑧𝑛 )

𝐷′ 𝐹 0
= det 𝑑𝑧1 ∧ ⋯ ∧ 𝑑𝑧𝑛 ∧ 𝑑𝑧1 ∧ ⋯ ∧ 𝑑𝑧𝑛
( 0 𝐷′ 𝐹 )
= | det 𝐷′ 𝐹 |2 𝑑𝑧1 ∧ ⋯ ∧ 𝑑𝑧𝑛 ∧ 𝑑𝑧1 ∧ ⋯ ∧ 𝑑𝑧𝑛 ,

and multiplying both sides by (−1)(𝑛−1)𝑛/2 (𝑖/2)𝑛 implies

𝐹 ∗ 𝜔𝑛 = | det 𝐷′ 𝐹 |2 𝜔𝑛 .

This shows that every biholomorphism between open subsets of ℂ𝑛 is orientation-

Now let 𝑀 be an 𝑛-dimensional complex manifold. Because every holomor-

preserving.

𝑀 satisfying (i) and (ii), it must be determined in the domain of each holomorphic
phic coordinate chart is a local biholomorphism, if there is to be an orientation of

chart by the pullback of 𝜔𝑛 under the coordinate map, and it is uniquely determined
by this property. We just need to verify that the orientations determined by different

Suppose two holomorphic charts (𝑈 , 𝜑) and (𝑉 , 𝜓) overlap. The transition

holomorphic charts agree.

function 𝜓 ∘ 𝜑−1 is a biholomorphism between open subsets of ℂ𝑛 , so the above

computation shows that (𝜓 ∘ 𝜑−1 )∗ 𝜔𝑛 = 𝑢 𝜔𝑛 , where 𝑢 is the positive smooth func-
tion | det 𝐷′ (𝜓 ∘ 𝜑−1 )|2 . Thus on 𝑈 ∩ 𝑉 we have

𝜓 ∗ 𝜔𝑛 = 𝜑∗ (𝜑−1 )∗ 𝜓 ∗ 𝜔𝑛
= 𝜑∗ ((𝜓 ∘ 𝜑−1 )∗ 𝜔𝑛 )
= 𝜑 ∗ (𝑢 𝜔 𝑛 )
= (𝑢 ∘ 𝜑)𝜑∗ 𝜔𝑛 .

Thus the 𝑛-forms determined by 𝜑 and 𝜓 are positive multiples of each other, so
they determine the same orientation on 𝑈 ∩ 𝑉 .

manifolds is orientation-preserving. Suppose 𝐹 ∶ 𝑀 → 𝑁 is a local biholomor-

Finally, we need to show that every local biholomorphism between complex

phism. Let 𝑝 ∈ 𝑀, and choose holomorphic charts (𝑈 , 𝜑) for 𝑀 and (𝑉 , 𝜓) for 𝑁

such that 𝑝 ∈ 𝑈 , 𝐹 (𝑈 ) ⊆ 𝑉 , and 𝐹 |𝑈 is a biholomorphism onto its image. Then
on 𝑈 ,
𝐹 = (𝜓 −1 ) ∘ (𝜓 ∘ 𝐹 ∘ 𝜑−1 ) ∘ (𝜑).

third by the way we have defined the orientations on 𝑁 and 𝑀, and the second
The three maps in parentheses above are all orientation-preserving: the first and

because it is a biholomorphism between open subsets of ℂ𝑛 . □

34 1. The Basics

Almost Complex Structures

To delve further into the interaction between a holomorphic structure and its under-

Let 𝑉 be an 𝑛-dimensional complex vector space, and let 𝑉ℝ be its underlying real
lying smooth structure, we introduce the following linear-algebraic construction.

vector space—the same set as 𝑉 , but considered only as a vector space over ℝ. Then
𝑉ℝ is a 2𝑛-dimensional real vector space. The fact that 𝑉 is a complex vector space
is encoded in the rule for multiplying vectors by 𝑖, which is the map 𝐽 ∶ 𝑉 → 𝑉
sending each vector 𝑣 to 𝑖𝑣. By ignoring the complex vector space structure, we
can also think of 𝐽 as a real-linear map 𝐽 ∶ 𝑉ℝ → 𝑉ℝ satisfying 𝐽 ∘ 𝐽 = − Id.
Now suppose 𝑉 is any vector space over ℝ. A complex structure on 𝑽 is a
real-linear endomorphism 𝐽 ∶ 𝑉 → 𝑉 satisfying 𝐽 ∘ 𝐽 = − Id.
Lemma 1.50. Suppose 𝑉 is a real vector space and 𝐽 is a complex structure on
𝑉 . Then the multiplication by complex scalars defined by (𝑎 + 𝑏𝑖)𝑣 = 𝑎𝑣 + 𝑏𝐽 𝑣,
together with the given vector addition operation, turns the set 𝑉 into a complex
vector space.

► Exercise 1.51. Prove this lemma by showing that complex multiplication is

associative and distributive.

To understand a complex structure 𝐽 on a vector space 𝑉 more deeply, we need

to look at its eigenvalues. The fact that 𝐽 ∘ 𝐽 = − Id means that every eigenvalue
𝜆 must satisfy 𝜆2 = −1. Thus 𝐽 has no real eigenvalues, and the only possible
complex eigenvalues are ±𝑖. To find eigenspaces, therefore, we must complexify
𝑉 and 𝐽 . Let 𝑉ℂ be the complexification of 𝑉 , and denote the complexification of
𝐽 by 𝐽 ∶ 𝑉ℂ → 𝑉ℂ . It still satisfies 𝐽 ∘ 𝐽 = − Id.
Proposition 1.52. If 𝐽 is a complex structure on the real vector space 𝑉 , then 𝑉ℂ
has a complete eigenspace decomposition of the form
𝑉ℂ = 𝑉 ′ ⊕ 𝑉 ″ ,
where 𝑉 ′ ⊆ 𝑉ℂ is the 𝑖-eigenspace of 𝐽 and 𝑉 ″ is the (−𝑖)-eigenspace. The
eigenspace decomposition of 𝑤 ∈ 𝑉ℂ is given by 𝑤 = 𝑤′ + 𝑤″ , where
(1.17) 𝑤′ = 21 (𝑤 − 𝑖𝐽 𝑤), 𝑤″ = 12 (𝑤 + 𝑖𝐽 𝑤).
If 𝑉 is finite-dimensional, then 𝑉 ′ and 𝑉 ″ have the same complex dimension.

Proof. Given 𝑤 ∈ 𝑉ℂ , define 𝑤′ , 𝑤″ ∈ 𝑉ℂ by (1.17). Simple computations show

that 𝐽 𝑤′ = 𝑖𝑤′ and 𝐽 𝑤″ = −𝑖𝑤″ . Because 𝑤 = 𝑤′ + 𝑤″ , this shows that
𝑉ℂ = 𝑉 ′ + 𝑉 ″ . On the other hand, a nonzero vector cannot be an eigenvector with
two different eigenvalues, so 𝑉 ′ ∩ 𝑉 ″ = {0}, which shows that the sum is direct.

(Exercise 1.39) is a bijective real-linear map from 𝑉ℂ to itself, and it interchanges

To see that the eigenspaces have the same dimension, note that conjugation
Almost Complex Structures 35

𝑉 ′ and 𝑉 ″ . Thus the underlying real spaces of 𝑉 ′ and 𝑉 ″ have the same real
dimension, and because the complex dimension is half the real dimension, 𝑉 ′ and
𝑉 ″ have the same complex dimension. □

Corollary 1.53. If a finite-dimensional real vector space admits a complex struc-

ture, then it is even-dimensional.

Proof. If 𝑉 admits a complex structure, the preceding proposition shows that 𝑉ℂ

of 𝑉ℂ is equal to the real dimension of 𝑉 .

is even-dimensional. The result follows from the fact that the complex dimension
□

Let us apply this construction to ℂ𝑛 with its standard complex structure. Let
(𝑋1 , … , 𝑋𝑛 ) denote the standard basis for ℂ𝑛 as a complex vector space, where
𝑋𝑗 = (0, … , 1, … , 0) with a 1 in the 𝑗th place. Let 𝑌𝑗 = 𝐽 𝑋𝑗 = (0, … , 𝑖, … , 0).
Then (𝑋1 , 𝑌1 , … , 𝑋𝑛 , 𝑌𝑛 ) is a basis over ℝ for the underlying real vector space
(ℂ𝑛 )ℝ , and 𝐽 satisfies 𝐽 𝑋𝑗 = 𝑌𝑗 , 𝐽 𝑌𝑗 = −𝑋𝑗 . From Proposition 1.52, we see
that the 𝑖-eigenspace (ℂ𝑛 )′ is spanned by (𝑍1 , … , 𝑍𝑛 ), where 𝑍𝑗 = 12 (𝑋𝑗 − 𝑖𝑌𝑗 ),
and (ℂ𝑛 )″ is spanned by (𝑍 1 , … , 𝑍 𝑛 ).
All of these constructions can be applied to vector bundles. If 𝐸 → 𝑀 is a
smooth real vector bundle, a complex structure on 𝑬 is a smooth bundle endomor-
phism 𝐽 ∶ 𝐸 → 𝐸 satisfying 𝐽 ∘ 𝐽 = − Id.
Consider the case of ℂ𝑛 as a smooth manifold. For each point 𝑝 ∈ ℂ𝑛 , using the
standard identification of 𝑇𝑝 ℂ𝑛 with (ℂ𝑛 )ℝ , we have the following correspondences:

𝜕 𝜕 𝜕
↔ 𝑋𝑗 , ↔ 𝑌𝑗 , ↔ 𝑍𝑗 .
𝜕𝑥𝑗 |𝑝 𝜕𝑦𝑗 |𝑝 𝜕𝑧𝑗 |𝑝

Thus the bundle 𝑇 ℂ𝑛 has a canonical complex structure 𝐽ℂ𝑛 , which satisfies
𝜕 𝜕 𝜕 𝜕
𝐽ℂ𝑛 = 𝑗, 𝐽ℂ𝑛 = − 𝑗.
𝜕𝑥 𝑗 𝜕𝑦 𝜕𝑦 𝑗 𝜕𝑥
The complexified tangent bundle 𝑇ℂ ℂ𝑛 splits as 𝑇ℂ ℂ𝑛 = 𝑇 ′ ℂ𝑛 ⊕ 𝑇 ″ ℂ𝑛 , with
𝑇 ′ ℂ𝑛 spanned by the complex vector fields 𝜕/𝜕𝑧1 , … , 𝜕/𝜕𝑧𝑛 , and 𝑇 ″ ℂ𝑛 spanned
by 𝜕/𝜕𝑧1 , … , 𝜕/𝜕𝑧𝑛 .
Lemma 1.54. For an open subset 𝑈 ⊆ ℂ𝑛 , a smooth function 𝐹 ∶ 𝑈 → ℂ𝑚 is
holomorphic if and only if the following relation holds for all 𝑝 ∈ 𝑈 :

(1.18) 𝐷𝐹 (𝑝) ∘ 𝐽ℂ𝑛 = 𝐽ℂ𝑚 ∘ 𝐷𝐹 (𝑝).

Proof. First suppose that (1.18) holds for all 𝑝 ∈ 𝑈 . After both sides are ex-
tended by complex linearity to act on complex vectors, the two expressions yield
the same result when applied to the elements of the complex coordinate frame
36 1. The Basics

{𝜕/𝜕𝑧𝑗 , 𝜕/𝜕𝑧𝑗 }. Using (1.12), we obtain

𝜕 𝜕
0 = 𝐷𝐹 (𝐽ℂ𝑛 ) − 𝐽ℂ𝑚 (𝐷𝐹 𝑗 )
𝜕𝑧 𝑗 𝜕𝑧
𝜕 𝜕
= 𝐷𝐹 (−𝑖 𝑗 ) − 𝐽ℂ𝑚 (𝐷𝐹 𝑗 )
𝜕𝑧 𝜕𝑧
𝜕𝐹 𝑘 𝜕 𝜕𝐹 𝑘 𝜕 𝜕𝐹 𝑘 𝜕 𝜕𝐹 𝑘 𝜕
= −𝑖 − 𝑖 − 𝐽 − 𝐽
𝜕𝑧𝑗 𝜕𝑤𝑘 𝜕𝑧𝑗 𝜕𝑤𝑘 ℂ
𝜕𝑧𝑗 𝜕𝑤𝑘 ℂ
𝜕𝑧𝑗 𝜕𝑤𝑘
𝑚 𝑚

𝜕𝐹 𝜕
𝑘
= −2𝑖 𝑗 .
𝜕𝑧 𝜕𝑤𝑘
This shows 𝜕𝐹 𝑘 /𝜕𝑧𝑗 ≡ 0 for all 𝑗, 𝑘, so 𝐹 is holomorphic.
Conversely, if 𝐹 is holomorphic, the computation above shows that both sides
of (1.18) yield the same result when applied to 𝜕/𝜕𝑧𝑗 , and conjugation shows that
the same is true when applied to 𝜕/𝜕𝑧𝑗 , using the fact that 𝜕𝐹 𝑘 /𝜕𝑧𝑗 = 𝜕𝐹 𝑘 /𝜕𝑧𝑗 = 0.
Since both sides are linear over 𝐶 ∞ (𝑀; ℂ), this shows the equation holds when
applied to arbitrary vector fields. □

Lemma 1.54 enables us to define a canonical complex structure on the tangent

bundle of every complex manifold.
Proposition 1.55. For every complex manifold 𝑀, there is a canonical complex
structure on 𝑇 𝑀, denoted by 𝐽𝑀 ∶ 𝑇 𝑀 → 𝑇 𝑀. If 𝑁 is another complex manifold
and 𝐹 ∶ 𝑀 → 𝑁 is a smooth map, then 𝐹 is holomorphic if and only if

(1.19) 𝐷𝐹 ∘ 𝐽𝑀 = 𝐽𝑁 ∘ 𝐷𝐹 .

Proof. Let 𝑛 be the complex dimension of 𝑀. We define 𝐽𝑀 as follows: given

𝑝 ∈ 𝑀, choose a holomorphic coordinate chart (𝑈 , 𝜑) on a neighborhood of 𝑝, and
define 𝐽𝑀 ∶ 𝑇 𝑀|𝑈 → 𝑇 𝑀|𝑈 by

(1.20) 𝐽𝑀 = 𝐷𝜑−1 ∘ 𝐽ℂ𝑛 ∘ 𝐷𝜑.

Wherever two holomorphic charts (𝑈 , 𝜑) and (𝑉 , 𝜓) overlap, the transition map 𝜓 ∘

𝜑−1 is a holomorphic map between open subsets of ℂ𝑛 , so its differential commutes
with 𝐽ℂ𝑛 by Lemma 1.54. Therefore,

𝐷𝜓 −1 ∘ 𝐽ℂ𝑛 ∘ 𝐷𝜓 = 𝐷𝜓 −1 ∘ 𝐽ℂ𝑛 ∘ (𝐷𝜓 ∘ 𝐷𝜑−1 ) ∘ 𝐷𝜑

= 𝐷𝜓 −1 ∘ (𝐷𝜓 ∘ 𝐷𝜑−1 ) ∘ 𝐽ℂ𝑛 ∘ 𝐷𝜑
= 𝐷𝜑−1 ∘ 𝐽ℂ𝑛 ∘ 𝐷𝜑,

so 𝐽𝑀 is well defined. The fact that it satisfies 𝐽𝑀 ∘ 𝐽𝑀 = − Id follows from the

corresponding fact for 𝐽ℂ𝑛 .
Almost Complex Structures 37

Now let 𝑁 be a complex 𝑚-manifold and 𝐹 ∶ 𝑀 → 𝑁 be a smooth map.

charts (𝑈 , 𝜑) for 𝑀 and (𝑉 , 𝜓) for 𝑁 such that 𝐹 (𝑈 ) ⊆ 𝑉 , and prove that the
Because (1.19) is a local statement, it suffices to choose arbitrary local holomorphic

restriction of 𝐹 to 𝑈 is holomorphic if and only if it satisfies (1.19) there. By

definition, 𝐹 is holomorphic on 𝑈 if and only if its coordinate representation 𝐹̂ =
𝜓 ∘ 𝐹 ∘ 𝜑−1 is holomorphic, which in turn is true if and only if 𝐷𝐹̂ ∘ 𝐽ℂ𝑛 = 𝐽ℂ𝑚 ∘ 𝐷𝐹̂
by Lemma 1.54. Using (1.20) for both 𝑀 and 𝑁, we compute

𝐷𝐹̂ ∘ 𝐽ℂ𝑛 − 𝐽ℂ𝑚 ∘ 𝐷𝐹̂ = 𝐷𝜓 ∘ 𝐷𝐹 ∘ 𝐷𝜑−1 ∘ 𝐽ℂ𝑛 − 𝐽ℂ𝑚 ∘ 𝐷𝜓 ∘ 𝐷𝐹 ∘ 𝐷𝜑−1

= 𝐷𝜓 ∘ 𝐷𝐹 ∘ 𝐽𝑀 ∘ 𝐷𝜑−1 − 𝐷𝜓 ∘ 𝐽𝑁 ∘ 𝐷𝐹 ∘ 𝐷𝜑−1
= 𝐷𝜓 ∘ (𝐷𝐹 ∘ 𝐽𝑀 − 𝐽𝑁 ∘ 𝐷𝐹 ) ∘ 𝐷𝜑−1 .

Since 𝐷𝜓 and 𝐷𝜑−1 are bundle isomorphisms, this last expression is zero if and
only if (1.18) holds, thus completing the proof. □

Proposition 1.56. Let 𝑀 be a complex manifold and let 𝐽𝑀 ∶ 𝑇 𝑀 → 𝑇 𝑀 be

the associated complex structure on 𝑇 𝑀. There are smooth subbundles 𝑇 ′ 𝑀,
𝑇 ″ 𝑀 ⊆ 𝑇ℂ 𝑀 whose fibers at each point are the 𝑖-eigenspace and (−𝑖)-eigenspace
of (the complexification of ) 𝐽𝑀 , respectively. The complexified tangent bundle
decomposes as a Whitney sum: 𝑇ℂ 𝑀 = 𝑇 ′ 𝑀 ⊕ 𝑇 ″ 𝑀. In terms of any local
holomorphic coordinates 𝑧𝑗 = 𝑥𝑗 + 𝑖𝑦 𝑗 , the complex vector fields 𝜕/𝜕𝑧𝑗 defined by
(1.8) form a local frame for 𝑇 ′ 𝑀; and the vector fields 𝜕/𝜕𝑧𝑗 form a local frame
for 𝑇 ″ 𝑀.

Proof. For each 𝑝 ∈ 𝑀, the space (𝑇𝑝 𝑀)ℂ has such a decomposition by Proposi-
tion 1.52. Suppose 𝑧𝑗 = 𝑥𝑗 +𝑖𝑦𝑗 are holomorphic local coordinates on 𝑀. Because
the endomorphism 𝐽𝑀 is defined by using the coordinate map to transport 𝐽ℂ𝑛 to
the manifold, it follows that the vector fields 𝜕/𝜕𝑧𝑗 provide a local frame for 𝑇 ′ 𝑀,
as do 𝜕/𝜕𝑧𝑗 for 𝑇 ″ 𝑀. Because both subbundles are spanned locally by smooth
vector fields, they are smooth. □

We call the bundles 𝑇 ′ 𝑀 and 𝑇 ″ 𝑀 the holomorphic tangent bundle and an-
tiholomorphic tangent bundle of 𝑴, respectively. The fibers 𝑇𝑝′ 𝑀 and 𝑇𝑝″ 𝑀 at
a point 𝑝 ∈ 𝑀 are called the holomorphic tangent space and antiholomorphic
tangent space at 𝒑, respectively.
The decomposition of 𝑇ℂ 𝑀 into holomorphic and antiholomorphic tangent

cobian of a holomorphic map. It follows from Proposition 1.55 that if 𝐹 ∶ 𝑀 → 𝑁

bundles allows us to give a coordinate-free interpretation to the holomorphic Ja-

is holomorphic, then 𝐷𝐹 (𝑇 ′ 𝑀) ⊆ 𝑇 ′ 𝑁. In local holomorphic coordinates (𝑧𝑗 ) for

𝑀 and (𝑤𝑘 ) for 𝑁, Corollary 1.46 shows that the restriction of 𝐷𝐹 (𝑝) to 𝑇𝑝′ 𝑀 is
represented by the holomorphic Jacobian matrix (𝜕𝐹 𝑘 (𝑝)/𝜕𝑧𝑗 ). Henceforth, we will
use the notation 𝐷′ 𝐹 (𝑝) and the term holomorphic Jacobian to refer either to this
38 1. The Basics

complex-linear map from 𝑇𝑝′ 𝑀 to 𝑇𝐹′ (𝑝) 𝑁 or to its matrix representation in local
holomorphic coordinates.
For a finite-dimensional real vector space with its natural smooth structure,
the tangent space at each point is canonically identified with the vector space itself
[LeeSM, Prop. 3.13]. The following proposition shows that there is a corresponding
identification for complex vector spaces.

Suppose 𝑉 is a finite-dimensional complex vector space with its standard holo-

Proposition 1.57 (Holomorphic Tangent Space to a Complex Vector Space).

morphic structure. For each 𝑎 ∈ 𝑉 , there is a canonical (basis-independent)

complex-linear isomorphism Φ𝑎 ∶ 𝑉 ≅ 𝑇𝑎′ 𝑉 . It is natural in the following sense:
if 𝐿 ∶ 𝑉 → 𝑊 is a complex-linear map between finite-dimensional complex vector
spaces, then the following diagram commutes for each 𝑎 ∈ 𝑉 :
Φ𝑎
𝑉 / 𝑇 ′𝑉
𝑎

(1.21) 𝐿 𝐷′ 𝐿(𝑎)
𝑊 / 𝑇′ 𝑊 .
Φ𝐿(𝑎) 𝐿(𝑎)
 

Proof. Given 𝑎, 𝑤 ∈ 𝑉 , let 𝜆𝑎,𝑤 ∶ ℂ → 𝑉 be the holomorphic map 𝜆𝑎,𝑤 (𝜏) =

𝑎 + 𝜏𝑤. We define Φ𝑎 ∶ 𝑉 → 𝑇𝑎′ 𝑉 by
𝜕
Φ𝑎 (𝑤) = 𝐷′ (𝜆𝑎,𝑤 )(0) ( | ) .
𝜕𝜏 0
The definition shows that this is independent of any choice of basis for 𝑉 . To see
that it satisfies the required conditions, choose any basis for 𝑉 and let (𝑧1 , … , 𝑧𝑛 ) be

lary 1.46 shows that Φ𝑎 has the coordinate representation

the corresponding linear coordinates. Then a simple computation based on Corol-

𝜕
Φ𝑎 (𝑤1 , … , 𝑤𝑛 ) = 𝑤𝑗 𝑗 | ,
𝜕𝑧 𝑎
which shows that it is a complex-linear isomorphism. If 𝑊 is another finite-
dimensional complex vector space and 𝐿 ∶ 𝑉 → 𝑊 is a complex-linear map, then
in terms of any linear coordinates (𝜁 1 , … , 𝜁 𝑚 ) for 𝑊 , we see that
𝜕
𝐷′ 𝐿(𝑎)(Φ𝑎 (𝑤1 , … , 𝑤𝑛 )) = 𝐿𝑘 𝑤𝑘 = Φ𝐿(𝑎) (𝐿(𝑤1 , … , 𝑤𝑛 )),
𝑗
𝜕𝜁 𝑗 |𝐿(𝑎)
which proves (1.21). □

For a complex manifold 𝑀, we have now introduced several different varieties

of tangent bundles: 𝑇 𝑀, 𝑇ℂ 𝑀, 𝑇 ′ 𝑀, and 𝑇 ″ 𝑀. In case you are not confused
enough already, we now define one more: 𝑇𝐽 𝑀 is the complex vector bundle with
the same total space as the ordinary tangent bundle 𝑇 𝑀, but endowed with the
complex vector space structure on fibers determined by 𝐽𝑀 as in Lemma 1.50.
Almost Complex Structures 39

Proposition 1.58. Let 𝑀 be a complex 𝑛-manifold. Then 𝑇𝐽 𝑀 is a smooth rank-𝑛

complex vector bundle over 𝑀. The complex vector bundles 𝑇𝐽 𝑀 and 𝑇 ′ 𝑀 are
isomorphic via the map 𝜉 ∶ 𝑇𝐽 𝑀 → 𝑇 ′ 𝑀 given by 𝜉(𝑣) = 𝑣 − 𝑖𝐽 𝑣.

Proof. Problem 1-7. □

For easy reference, here is a summary of all of these bundles. Suppose 𝑀 is a

complex 𝑛-manifold.
• 𝑻 𝑴: The ordinary tangent bundle of the smooth manifold 𝑀. It is a real
vector bundle of rank 2𝑛.
• 𝑻ℂ 𝑴: The complexified tangent bundle, a complex vector bundle of rank
2𝑛.
• 𝑻 ′ 𝑴: The holomorphic tangent bundle, a complex rank-𝑛 vector subbun-
dle of 𝑇ℂ 𝑀. Its fiber at each point is the 𝑖-eigenspace of 𝐽𝑀 .
• 𝑻 ″ 𝑴: The antiholomorphic tangent bundle, another complex rank-𝑛 vec-
tor subbundle of 𝑇ℂ 𝑀, whose fibers are (−𝑖)-eigenspaces of 𝐽𝑀 .
• 𝑻𝑱 𝑴: The ordinary tangent bundle of 𝑀 equipped with the complex
structure 𝐽𝑀 , which turns it into a complex vector bundle of rank 𝑛.
Now suppose 𝑀 is an arbitrary smooth manifold. It makes sense to ask whether
there is a complex structure on 𝑇 𝑀, that is, a smooth bundle endomorphism
𝐽 ∶ 𝑇 𝑀 → 𝑇 𝑀 satisfying 𝐽 ∘ 𝐽 = − Id. The existence of such an endomor-
phism is a necessary condition for the existence of a holomorphic structure on 𝑀,

tangent bundle is endowed with such a complex structure 𝐽 is called an almost

but it is not sufficient, as we will see below. For this reason, a manifold whose

complex manifold, and 𝐽 is called an almost complex structure on 𝑴. Note the

potentially confusing shift in terminology: a complex structure on 𝑇 𝑀 is called an
almost complex structure on 𝑀 (to distinguish it from the traditional use of “com-
plex structure on 𝑀” to denote what we are calling a holomorphic structure).
Proposition 1.55 shows that a holomorphic structure on a manifold 𝑀 deter-
mines an almost complex structure on 𝑀 (that is, a complex structure on 𝑇 𝑀).

structure 𝐽 on a smooth manifold 𝑀, is there a holomorphic structure for which

The question naturally arises whether the reverse is true: Given an almost complex

𝐽 is the canonical almost complex structure as described in Proposition 1.55? In

general, the answer is no, because there is a nontrivial necessary condition, as a
consequence of the next proposition.
Proposition 1.59. Suppose 𝑀 is a complex manifold and 𝑉 , 𝑊 ∈ Γ(𝑇 ′ 𝑀). Then
[𝑉 , 𝑊 ] ∈ Γ(𝑇 ′ 𝑀).

𝜕 𝜕
Proof. In local holomorphic coordinates, we can write

𝑉 = 𝑉𝑗 𝑗, 𝑊 = 𝑊 𝑘 𝑘,
𝜕𝑧 𝜕𝑧
40 1. The Basics

and therefore,
𝜕𝑊 𝑘 𝜕 𝑘 𝜕𝑉
𝑗
𝜕
[𝑉 , 𝑊 ] = 𝑉 𝑗 − 𝑊 .
( 𝜕𝑧𝑗 ) 𝜕𝑧𝑘 ( 𝜕𝑧𝑘 ) 𝜕𝑧𝑗
This last expression takes its values in 𝑇 ′ 𝑀. □

For almost complex structures, it makes sense to ask if the same result holds,
by virtue of the following lemma.
Lemma 1.60. Suppose 𝑀 is a smooth 2𝑛-manifold endowed with an almost com-
plex structure 𝐽 . Then there are smooth rank-𝑛 complex subbundles 𝑇 ′ 𝑀, 𝑇 ″ 𝑀 ⊆
𝑇ℂ 𝑀 whose fibers are the 𝑖-eigenspaces and (−𝑖)-eigenspaces of 𝐽 , respectively,
such that 𝑇ℂ 𝑀 = 𝑇 ′ 𝑀 ⊕ 𝑇 ″ 𝑀.

► Exercise 1.61. Prove this lemma.

An almost complex structure on a smooth manifold 𝑀 is said to be integrable if

whenever 𝑉 , 𝑊 are smooth sections of 𝑇 ′ 𝑀, then [𝑉 , 𝑊 ] is also a section of 𝑇 ′ 𝑀.
Every almost complex structure on a 2-dimensional real manifold is integrable (see
Problem 1-8), but in higher dimensions integrability is a nontrivial condition, as
Problems 1-11 and 1-13 illustrate.
The integrability condition looks formally similar to the condition of involu-
tivity for a distribution (subbundle of the tangent bundle) on a smooth manifold,

foliation (see [LeeSM, Chap. 19]). But there is no foliation associated with 𝑇 ′ 𝑀
which is a necessary and sufficient condition for the distribution to be tangent to a

because it is not a subbundle of the (real) tangent bundle of 𝑀.

The following corollary is an immediate consequence of Proposition 1.59.
Corollary 1.62. If 𝐽 is an almost complex structure on a smooth manifold, a neces-
sary condition for 𝐽 to be the canonical almost complex structure associated with
a holomorphic structure is that 𝐽 be integrable. □
The following important converse was proved in 1957 by August Newlander
and Louis Nirenberg, showing that integrability is also sufficient.
Theorem 1.63 (Newlander–Nirenberg). If an almost complex structure on a
smooth manifold is integrable, then it arises from a holomorphic structure.
We will neither prove nor use this theorem (except in Example 1.64 and Prob-
lem 7-10 below, which are not essential to our main story). There are several known
proofs, all based on deep results from the theory of partial differential equations.
Two different proofs can be found in [Nir73] and [Hör90].
Not every smooth manifold admits an almost complex structure. Two simple
requirements are that the manifold must be even-dimensional (Cor. 1.53) and ori-
entable (Problem 1-9). In two real dimensions, these conditions are sufficient, as
Problems 41

Example 1.64 below will show. But in higher dimensions, there are other topo-

complex structures are 𝕊2 (by Example 1.64 below) and 𝕊6 (by Problem 1-13). The
logical obstructions that are not so easily described. Two spheres that admit almost

structure on 𝕊2 is integrable, and turns it into a complex manifold biholomorphic to

ℂℙ1 (see Problem 2-4). The known structure on 𝕊6 is not integrable, and it is not
known whether 𝕊6 carries a holomorphic structure. It was proved by Armand Borel
and Jean-Pierre Serre in 1953 [BS53] that 𝕊2 and 𝕊6 are the only spheres that carry
almost complex structures; so no other spheres can be made into complex mani-
folds. A modern proof of this fact can be found in [May99, p. 208]. Since our main
concern is to study complex manifolds, which already come equipped with canon-
ical almost complex structures, we do not pursue the general question of existence
of almost complex structures any further.
Example 1.64 (Holomorphic Structures on 2-Manifolds). Suppose 𝑀 is an ori-
entable real 2-manifold. We can always endow 𝑀 with a Riemannian metric and
an orientation. With that data, we can define an almost complex structure on 𝑀 by
letting 𝐽 be “counterclockwise rotation by 90∘ .” More precisely, for each nonzero
𝑣 ∈ 𝑇𝑝 𝑀, we let 𝐽 𝑣 be the unique vector 𝑤 such that ⟨𝑣, 𝑤⟩ = 0, |𝑤| = |𝑣|, and
(𝑣, 𝑤) is an oriented basis for 𝑇𝑝 𝑀. If (𝑏1 , 𝑏2 ) is any smooth oriented orthonor-
mal local frame, then we have 𝐽 𝑏1 = 𝑏2 and 𝐽 𝑏2 = −𝑏1 , which shows that 𝐽 is
smooth. This almost complex structure is integrable by the result of Problem 1-

Thus every orientable smooth real 2-manifold can be given a holomorphic struc-
8, so it arises from a holomorphic structure by the Newlander–Nirenberg theorem.

ture. A real 2-manifold endowed with a particular holomorphic structure is called

complex curve (1-dimensional complex manifold), while a complex surface is a 2-

a Riemann surface. Be careful about the terminology: a Riemann surface is a

dimensional complex manifold. (It is possible for the same real 2-manifold to have
different holomorphic structures that are not biholomorphic to each other, however;
see Problem 1-4.) //

Problems
1-1. With 𝐺 ⊆ GL(3, ℂ) as in Example 1.20, let Γ ⊆ 𝐺 be the subgroup con-
sisting of matrices whose entries are Gaussian integers. Prove that Γ is
cocompact by showing that every coset in 𝐺/Γ has at least one represen-
tative lying in the unit cube [0, 1]6 ⊆ ℂ3 .
1-2. Suppose 𝑈 ⊆ ℂ𝑛 is open and 𝑓 ∶ 𝑈 → ℂ is a holomorphic function
that is nonzero on 𝑈 ∖ 𝑆, where 𝑆 ⊆ ℂ𝑛 is a complex-linear subspace of
codimension at least 2. Show that 𝑓 is nonzero everywhere in 𝑈 .
1-3. Prove that every 1-dimensional Hopf manifold is biholomorphic to a com-
plex torus ℂ/Λ, and determine an explicit lattice Λ.
42 1. The Basics

1-4. For any two vectors 𝑣, 𝑤 ∈ ℂ that are linearly independent over ℝ, let
𝑇𝑣,𝑤 = ℂ/Λ(𝑣, 𝑤) denote the 1-dimensional complex torus obtained as a
quotient of ℂ by the lattice Λ(𝑣, 𝑤) generated by 𝑣 and 𝑤.
(a) For any such 𝑣, 𝑤, show that there exists 𝜏 ∈ ℂ with Im 𝜏 > 0 such
that 𝑇𝑣,𝑤 is biholomorphic to 𝑇1,𝜏 .
(b) Let SL(2, ℤ) denote the group of integer matrices with determinant
1. Suppose 𝜏, 𝜏 ′ ∈ ℂ satisfy Im 𝜏 > 0 and Im 𝜏 ′ > 0. Show that 𝑇1,𝜏
is biholomorphic to 𝑇1,𝜏 ′ if and only if there exists ( 𝑎𝑐 𝑑𝑏 ) ∈ SL(2, ℤ)
such that 𝜏 ′ = (𝑎𝜏 + 𝑏)/(𝑐𝜏 + 𝑑). [Hint: Show that any biholomor-
phism 𝑇1,𝜏 → 𝑇1,𝜏 ′ lifts to an automorphism of ℂ.]

Grassmannian G𝑘 (ℂ𝑛 ) by 𝐴 ⋅ 𝑆 = 𝐴(𝑆) for 𝐴 ∈ U(𝑛) and 𝑆 ⊆ ℂ𝑛 a

1-5. Show that the Lie group U(𝑛) acts continuously and transitively on the

subspace of dimension 𝑘. Use this to show that G𝑘 (ℂ𝑛 ) is compact for

every 𝑘 and 𝑛.
1-6. Suppose 𝐸 → 𝑀 is a complex vector bundle. Show that there exists a

𝑐 ∶ 𝐸 → 𝐸 satisfying 𝑐 ∘ 𝑐 = Id, if and only if 𝐸 is isomorphic (over ℂ)

conjugation operator, that is, a conjugate-linear bundle homomorphism

to the complexification of a real bundle.

1-7. Prove Proposition 1.58 (𝑇𝐽 𝑀 is a smooth complex vector bundle isomor-
phic to 𝑇 ′ 𝑀).
1-8. Prove that every almost complex structure on a real 2-manifold is inte-
grable.
1-9. Suppose 𝑀 is a smooth manifold that admits an almost complex structure.
Prove that 𝑀 is orientable.
1-10. Let 𝑀 be a smooth manifold and 𝐽 be an almost complex structure on 𝑀.
Define a map 𝑁 ∶ Γ(𝑇 𝑀) × Γ(𝑇 𝑀) → Γ(𝑇 𝑀) by
𝑁(𝑋, 𝑌 ) = [𝐽 𝑋, 𝐽 𝑌 ] − [𝑋, 𝑌 ] − 𝐽 [𝑋, 𝐽 𝑌 ] − 𝐽 [𝐽 𝑋, 𝑌 ].
(a) Show that 𝑁 is bilinear over 𝐶 ∞ (𝑀), and therefore defines a (1, 2)-
tensor field on 𝑀, called the Nijenhuis tensor of 𝑱 .
(b) Show that 𝐽 is integrable if and only if 𝑁 ≡ 0. [Hint: Extend 𝑁 to
act on complex vector fields, and take 𝑋 and 𝑌 to be smooth sections
of 𝑇 ′ 𝑀 or 𝑇 ″ 𝑀.]
1-11. For 𝑛 ≥ 2, define an almost complex structure on ℂ𝑛 as follows:
𝜕 𝜕 𝜕 1 𝜕
𝐽 = (1 + (𝑥2 )2 ) 1 , 𝐽 =− ,
𝜕𝑥 1 𝜕𝑦 𝜕𝑦 1 (1 + (𝑥 ) ) 𝜕𝑥1
2 2

𝜕 𝜕 𝜕 𝜕
𝐽 𝑘 = 𝑘, 𝐽 𝑘 = − 𝑘, 𝑘 = 2, … , 𝑛.
𝜕𝑥 𝜕𝑦 𝜕𝑦 𝜕𝑥
Show that 𝐽 is not integrable.
Problems 43

1-12. Let 𝑀 be a 2𝑛-dimensional smooth manifold. Suppose 𝜁 is a smooth

closed complex 𝑛-form on 𝑀 that is locally decomposable (i.e., can lo-
cally be written as a wedge product of complex 1-forms), and satisfies
𝜁 ∧ 𝜁 ≠ 0 everywhere on 𝑀. Show that there is a unique integrable al-
most complex structure on 𝑀 for which 𝑇 ′ 𝑀 = {𝑣 ∈ 𝑇ℂ 𝑀 ∶ 𝑣 ⅃ 𝜁 = 0}
(where ⅃ denotes interior multiplication with a vector field, defined by
(𝑣 ⅃ 𝜁)(… ) = 𝜁(𝑣, … ); see [LeeSM, p. 358]).
1-13. AN ALMOST COMPLEX STRUCTURE ON 𝕊6 : Let 𝕆 denote the algebra of oc-
tonions, which is an 8-dimensional nonassociative algebra over ℝ defined
as follows. Start with the quaternions, the 4-dimensional associative al-
gebra ℍ over ℝ with basis (𝟙, 𝕚, 𝕛, 𝕜) and bilinear multiplication defined
by

𝕚2 = 𝕛2 = 𝕜2 = −𝟙, 𝟙𝑞 = 𝑞𝟙 = 𝑞 for all 𝑞 ∈ ℍ,

𝕚𝕛 = −𝕛𝕚 = 𝕜, 𝕛𝕜 = −𝕜𝕛 = 𝕚, 𝕜𝕚 = −𝕚𝕜 = 𝕛.

Then define 𝕆 = ℍ × ℍ, with the bilinear product defined by

(𝑝, 𝑞)(𝑟, 𝑠) = (𝑝𝑟 − 𝑠𝑞 ∗ , 𝑝∗ 𝑠 + 𝑟𝑞),

where the conjugate of a quaternion is

(𝑤𝟙 + 𝑥𝕚 + 𝑦𝕛 + 𝑧𝕜)∗ = 𝑤𝟙 − 𝑥𝕚 − 𝑦𝕛 − 𝑧𝕜.

Define the conjugate of an octonion 𝑃 = (𝑝, 𝑞) by 𝑃 ∗ = (𝑝∗ , −𝑞). Let

ℝ = {𝑃 ∈ 𝕆 ∶ 𝑃 ∗ = 𝑃 } denote the set of real octonions, identified with
the real numbers in the natural way, and 𝔼 = {𝑃 ∈ 𝕆 ∶ 𝑃 ∗ = −𝑃 } the
set of imaginary octonions. Define an inner product on 𝕆 by ⟨𝑃 , 𝑄⟩ =
1
2
(𝑃 ∗ 𝑄 + 𝑄∗ 𝑃 ) ∈ ℝ. Let 𝕊 = {𝑃 ∈ 𝔼 ∶ |𝑃 | = 1} be the unit sphere in 𝔼,
and for each 𝑃 ∈ 𝕊, define a map 𝐽𝑃 ∶ 𝑇𝑃 𝕊 → 𝕆 by 𝐽𝑃 (𝑄) = 𝑄𝑃 , where
we identify 𝑇𝑃 𝕊 with the real-linear subspace 𝑃 ⟂ ∩ 𝔼 ⊆ 𝕆. Although the
multiplication in 𝕆 is not associative, it is the case that (𝑃 𝑄)∗ = 𝑄∗ 𝑃 ∗
and (𝑃 𝑄)𝑃 = 𝑃 (𝑄𝑃 ) for all 𝑃 , 𝑄 ∈ 𝕆, and you may use these facts

(a) Show that 𝐽𝑃 maps 𝑇𝑃 𝕊 to itself, and defines an almost complex

without proof. (See also Problem 8-7 in [LeeSM].)

structure on 𝕊.

[Remark: It is still unknown whether 𝕊6 admits an integrable almost com-

(b) Show that this almost complex structure is not integrable.

plex structure. Many well-known and respected mathematicians have

written papers purporting to answer this question one way or the other,
but all the proofs have been found to be wrong or incomplete.]
44 1. The Basics

1-14. Let (𝑀, 𝑔) and (𝑁, ℎ) be Riemannian manifolds of the same dimension.
A smooth map 𝐹 ∶ 𝑀 → 𝑁 is said to be conformal if 𝐹 ∗ ℎ = 𝜆𝑔 for
some smooth, positive function 𝜆 on 𝑀.
(a) Suppose (𝑀, 𝑔) and (𝑁, ℎ) are oriented Riemannian 2-manifolds,
and give 𝑀 and 𝑁 the holomorphic structures described in Exam-
ple 1.64. Suppose 𝐹 ∶ 𝑀 → 𝑁 is a local diffeomorphism. Show
that 𝐹 is holomorphic if and only if it is conformal and orientation-

(b) Give examples of diffeomorphisms 𝐹 , 𝐺 ∶ ℂ2 → ℂ2 such that 𝐹 is

preserving.

holomorphic but not conformal, and 𝐺 is conformal and orientation-

preserving but not holomorphic.
 10.1090/gsm/244/02

Chapter 2

Complex Submanifolds

In this chapter, we introduce tools for studying complex submanifolds, which will
yield a rich new source of examples of complex manifolds.

Variations on the Inverse Function Theorem

In smooth manifold theory, the primary technical tools for constructing new smooth
manifolds out of old ones are the inverse function theorem and its friends, the im-
plicit function theorem and the rank theorem (see [LeeSM, Chaps. 4 and 5]). All
of these generalize straightforwardly to the holomorphic category.
Theorem 2.1 (Holomorphic Inverse Function Theorem). Suppose 𝑀 and 𝑁 are
complex 𝑛-manifolds, 𝐹 ∶ 𝑀 → 𝑁 is holomorphic, and the holomorphic Jacobian
𝐷′ 𝐹 (𝑝) is nonsingular for some 𝑝 ∈ 𝑀. Then there exist connected neighborhoods
𝑈0 of 𝑝 and 𝑉0 of 𝐹 (𝑝) such that 𝐹 restricts to a biholomorphism from 𝑈0 to 𝑉0 .

Proof. The hypothesis implies that 𝐷𝐹 (𝑝) is nonsingular, so by the ordinary inverse
function theorem there are connected neighborhoods 𝑈0 of 𝑝 and 𝑉0 of 𝐹 (𝑝) such
that 𝐹 |𝑈0 ∶ 𝑈0 → 𝑉0 is a diffeomorphism. We already know 𝐹 is holomorphic, so
it remains only to show that 𝐹 −1 is also.
By choosing local holomorphic coordinates on 𝑈0 and 𝑉0 (after shrinking both
neighborhoods if necessary) and replacing 𝐹 by its coordinate representation, we
can reduce the problem to the case in which 𝐹 is a holomorphic diffeomorphism
between open subsets of ℂ𝑛 , and we can use coordinates (𝑧1 , … , 𝑧𝑛 ) for both the
domain and codomain. Let 𝐺 = 𝐹 −1 , and consider the 𝑙th coordinate function 𝐺𝑙 .
It satisfies 𝐺𝑙 ∘ 𝐹 (𝑧) = 𝑧𝑙 . From the chain rule (Prop. 1.47), we see that
𝜕𝑧𝑙 𝜕(𝐺𝑙 ∘ 𝐹 ) 𝜕𝐺𝑙 𝜕𝐹 𝑘 𝜕𝐺𝑙 𝜕𝐹 𝑘
0= = = 𝑘 + 𝑘 .
𝜕𝑧𝑗 𝜕𝑧𝑗 𝜕𝑧 𝜕𝑧𝑗 𝜕𝑧 𝜕𝑧𝑗

45
46 2. Complex Submanifolds

The first term on the right-hand side vanishes because 𝐹 is holomorphic. Because
the matrix (𝜕𝐹 𝑘 𝜕𝑧𝑘 ) is invertible (it is the conjugate of the holomorphic Jacobian
of 𝐹 ), this implies that 𝜕𝐺𝑙 /𝜕𝑧𝑘 ≡ 0. Since this is true for all 𝑙 and 𝑘, it follows that
𝐺 is holomorphic.
Theorem 2.2 (Holomorphic Implicit Function Theorem). Let 𝑈 ⊆ ℂ𝑛 × ℂ𝑚 be
□

an open subset, and denote the standard holomorphic coordinates there by (𝑧, 𝑤) =
(𝑧1 , … , 𝑧𝑛 , 𝑤1 , … , 𝑤𝑚 ). Suppose Φ ∶ 𝑈 → ℂ𝑚 is a holomorphic map, and the
𝑚 × 𝑚 matrix (𝜕Φ𝑗 /𝜕𝑤𝑘 ) is nonsingular at some (𝑎, 𝑏) ∈ 𝑈 . Let 𝑐 = Φ(𝑎, 𝑏). Then
there exist neighborhoods 𝑉0 of 𝑎 in ℂ𝑛 and 𝑊0 of 𝑏 in ℂ𝑚 and a holomorphic
function 𝐹 ∶ 𝑉0 → 𝑊0 such that Φ−1 (𝑐) ∩ (𝑉0 × 𝑊0 ) is the graph of 𝐹 :
Φ−1 (𝑐) ∩ (𝑉0 × 𝑊0 ) = {(𝑧, 𝑤) ∈ 𝑉0 × 𝑊0 ∶ 𝑤 = 𝐹 (𝑧)}.

Proof. By shrinking 𝑈 if necessary, we may assume that the matrix (𝜕Φ𝑗 /𝜕𝑤𝑘 ) is
nonsingular on all of 𝑈 . The hypothesis implies that the holomorphic Jacobian of
the map 𝑤 ↦ Φ(𝑎, 𝑤) is nonsingular at 𝑤 = 𝑏, so Corollary 1.46 shows that the

function theorem then implies the existence of neighborhoods 𝑉0 and 𝑊0 and a

total derivative of this map is also nonsingular. The smooth version of the implicit

smooth function 𝐹 ∶ 𝑉0 → 𝑊0 such that Φ−1 (𝑐) ∩ (𝑉0 × 𝑊0 ) is the graph of 𝐹 . It

remains only to show that 𝐹 is holomorphic.
The fact that Φ(𝑧, 𝐹 (𝑧)) ≡ 𝑐 for 𝑧 ∈ 𝑉0 , together with the chain rule, implies
that for each 𝑙 and 𝑗,
𝜕 𝑙 𝜕Φ𝑙 𝜕Φ𝑙 𝜕𝐹 𝑘 𝜕Φ𝑙 𝜕𝐹 𝑘
0= Φ (𝑧, 𝐹 (𝑧)) = + + .
𝜕𝑧𝑗 𝜕𝑧𝑗 𝜕𝑤𝑘 𝜕𝑧𝑗 𝜕𝑤𝑘 𝜕𝑧𝑗
The first term and the last term on the right-hand side are zero because Φ is holo-
morphic. As in the previous proof, the invertibility of the matrix (𝜕Φ𝑙 /𝜕𝑤𝑘 ) on 𝑈
ensures that 𝜕𝐹 𝑘 /𝜕𝑧𝑗 ≡ 0.
Theorem 2.3 (Holomorphic Rank Theorem). Suppose 𝑀 and 𝑁 are complex
□

manifolds of dimensions 𝑚 and 𝑛, respectively, and 𝐹 ∶ 𝑀 → 𝑁 is a holomorphic

map whose holomorphic Jacobian has constant rank 𝑟. For each 𝑝 ∈ 𝑀 there exist
holomorphic charts (𝑈 , 𝜑) for 𝑀 centered at 𝑝 and (𝑉 , 𝜓) for 𝑁 centered at 𝐹 (𝑝)
such that 𝐹 (𝑈 ) ⊆ 𝑉 , in which 𝐹 has a coordinate representation of the form
𝐹̂(𝑧1 , … , 𝑧𝑟 , 𝑧𝑟+1 , … , 𝑧𝑚 ) = (𝑧1 , … , 𝑧𝑟 , 0, … , 0).

► Exercise 2.4. Prove this theorem by verifying that the proof of the ordinary
rank theorem [LeeSM, Thm. 4.12] goes through essentially unchanged with the
holomorphic inverse function theorem substituted for its smooth counterpart.

A holomorphic map between complex manifolds is called a submersion, im-

mersion, or embedding if it has the corresponding property when considered as a
smooth map between smooth manifolds: a submersion has surjective total deriv-
ative everywhere, an immersion has injective total derivative everywhere, and an
Variations on the Inverse Function Theorem 47

embedding is an immersion that is also a homeomorphism onto its image in the

subspace topology. Because the rank of the total derivative of a holomorphic map

in terms of the latter; in particular, a holomorphic map 𝐹 ∶ 𝑀 → 𝑁 between com-

is twice that of its holomorphic Jacobian, these properties can also be characterized

plex manifolds is a submersion if and only if 𝐷′ 𝐹 has rank equal to the complex
dimension of 𝑁 everywhere, and is an immersion if and only if 𝐷′ 𝐹 has rank equal
to the complex dimension of 𝑀 everywhere.
For any continuous map 𝐹 ∶ 𝑀 → 𝑁 between topological spaces, recall that
a local section of 𝑭 is a continuous map 𝜎 ∶ 𝑈 → 𝑀 defined on an open subset
𝑈 ⊆ 𝑁 and satisfying 𝐹 ∘ 𝜎 = Id𝑈 .
Corollary 2.5 (Holomorphic Local Section Theorem). Suppose 𝜋 ∶ 𝑀 → 𝑁 is
a holomorphic submersion. Then every point of 𝑀 is in the image of a holomorphic
local section of 𝜋.

Proof. For any 𝑝 ∈ 𝑀, by the rank theorem we can choose holomorphic co-
ordinates centered at 𝑝 and 𝜋(𝑝) in which the coordinate representation of 𝜋 is
𝜋̂(𝑧1 , … , 𝑧𝑛 ) = (𝑧1 , … , 𝑧𝑟 ). The map 𝜎(𝑧1 , … , 𝑧𝑟 ) = (𝑧1 , … , 𝑧𝑟 , 0, … , 0) is a holo-
morphic local section of 𝜋 sending 𝜋(𝑝) to 𝑝. □

sions). Suppose 𝑀, 𝑁, and 𝑃 are complex manifolds. If 𝜋 ∶ 𝑀 → 𝑁 is a surjec-

Corollary 2.6 (Characteristic Property of Surjective Holomorphic Submer-

tive holomorphic submersion, then a map 𝐹 ∶ 𝑁 → 𝑃 is holomorphic if and only

if 𝐹 ∘ 𝜋 is holomorphic.

are holomorphic. For the “if” part, suppose 𝐹 ∘ 𝜋 is holomorphic. Given 𝑞 ∈ 𝑁,

Proof. The “only if” claim is just the fact that compositions of holomorphic maps

we can find a local holomorphic section 𝜎 of 𝜋 defined on a neighborhood 𝑈 of 𝑞,

and then 𝐹 |𝑈 = (𝐹 ∘ 𝜋) ∘ 𝜎, which is a composition of holomorphic maps.
Corollary 2.7 (Passing to a Holomorphic Quotient). Suppose 𝑀, 𝑁, and 𝑃 are
□

complex manifolds and 𝜋 ∶ 𝑀 → 𝑁 is a surjective holomorphic submersion. If

𝐹 ∶ 𝑀 → 𝑃 is a holomorphic map that is constant on the fibers of 𝜋, then there is
a unique holomorphic map 𝐹̃ ∶ 𝑁 → 𝑃 such that 𝐹̃ ∘ 𝜋 = 𝐹 .

► Exercise 2.8. Prove this corollary.

Example 2.9 (Projective Transformations of ℂℙ𝒏 ). Suppose 𝐴 ∶ ℂ𝑛+1 → ℂ𝑛+1

is an invertible complex-linear map. Because 𝐴(𝜆𝑧) = 𝜆𝐴(𝑧) for 𝜆 ∈ ℂ and
𝑧 ∈ ℂ𝑛+1 , Corollary 2.7 applied to 𝜋 ∘ 𝐴 ∶ ℂ𝑛+1 ∖ {0} → ℂℙ𝑛 shows that 𝐴 de-
scends to a holomorphic map 𝐴̃ ∶ ℂℙ𝑛 → ℂℙ𝑛 defined by 𝐴([𝑧]) ̃ = [𝐴𝑧]. Since
̃ ̃ it follows that 𝐴̃ is an automorphism, called a projective
𝐴−1 is an inverse for 𝐴,
transformation. In fact, as we will see eventually, every automorphism of ℂℙ𝑛 is
a projective transformation. (See Problem 2-9 for the case 𝑛 = 1, and Problem 9-9
for higher dimensions.) //
48 2. Complex Submanifolds

► Exercise 2.10. Let 𝑉 be an 𝑛-dimensional complex vector space and let ℙ(𝑉 )
be its projectivization (see Example 1.11). Show that the bijections ℂℙ𝑛 → ℙ(𝑉 )
determined by any two different bases of 𝑉 differ by a projective transformation
of ℂℙ𝑛 , and therefore the complex manifold structure of ℙ(𝑉 ) is independent of
the choice of basis.

Complex Submanifolds
If 𝑀 is a complex manifold, an (embedded) complex submanifold of 𝑴 is a subset
𝑆 ⊆ 𝑀 that is a topological manifold in the subspace topology, and is endowed
with a holomorphic structure such that the inclusion 𝑆 ↪ 𝑀 is a holomorphic
embedding. The (complex) codimension of 𝑺 is dimℂ 𝑀 − dimℂ 𝑆. A complex
hypersurface is a complex submanifold of complex codimension 1. (Note that al-
gebraic geometers, such as [Har77, GH94], typically define hypersurfaces more
generally to include subsets with certain kinds of singularities; but for us a hyper-
surface will always mean a complex submanifold.) It is also possible to define im-
mersed complex submanifolds analogously to immersed smooth ones; but we will
not be making use of the immersed case, so we assume henceforth without further
comment that all complex submanifolds are embedded.
For a smooth submanifold of a smooth manifold, we can always canonically
identify the tangent space to the submanifold with a subspace of the ambient tan-

of complex submanifolds. Suppose 𝑀 is a complex manifold and 𝑆 ⊆ 𝑀 is

gent space. There is an analogous identification for holomorphic tangent spaces

an embedded complex submanifold, and let 𝜄 ∶ 𝑆 ↪ 𝑀 be the inclusion map.

Since 𝜄 is a holomorphic embedding, for each 𝑝 ∈ 𝑆 the holomorphic Jacobian
𝐷′ 𝜄(𝑝) ∶ 𝑇𝑝′ 𝑆 → 𝑇𝑝′ 𝑀 is an injective complex-linear map. Using this map, we can
identify 𝑇𝑝′ 𝑆 as a complex-linear subspace of 𝑇𝑝′ 𝑀.
Suppose (𝑈 , 𝜑) is a holomorphic coordinate chart for 𝑀 and 𝐾 ⊆ ℂ𝑛 is a
complex-linear subspace of dimension 𝑘. The subset 𝜑−1 (𝐾) is called a holomor-
phic 𝒌-slice of 𝑼 .

Proposition 2.11 (Holomorphic Slice Criterion). Suppose 𝑀 is a complex 𝑛-

manifold and 𝑆 is a subset of 𝑀 with the subspace topology. Then 𝑆 has a unique

if and only if for some fixed 𝑘, each 𝑝 ∈ 𝑆 is contained in the domain of a holo-
complex manifold structure that makes it into an embedded complex submanifold

morphic coordinate chart (𝑈 , 𝜑) in which 𝑆 ∩ 𝑈 is a holomorphic 𝑘-slice (called

a holomorphic slice chart for 𝑺).

► Exercise 2.12. Prove this proposition by verifying that the analogous proof for
smooth manifolds [LeeSM, Thms. 5.8 and 5.31] carries through with the holomor-
phic rank theorem in place of its smooth analogue.
Complex Submanifolds 49

Corollary 2.13. Suppose 𝑀 is a complex 𝑛-manifold and 𝑆 ⊆ 𝑀. Then 𝑆 is a

complex 𝑘-submanifold if and only if each 𝑝 ∈ 𝑆 has a neighborhood 𝑈 in 𝑀 such
that 𝑆 ∩ 𝑈 is the zero set of a holomorphic submersion 𝐹 ∶ 𝑈 → ℂ𝑛−𝑘 (called a
local defining function for 𝑺).

► Exercise 2.14. Prove this corollary. [Hint: Start by showing that the subspace
𝐾 in the definition of holomorphic slice charts can always be taken to be the sub-
space defined by 𝑧𝑘+1 = ⋯ = 𝑧𝑛 = 0.]

Proposition 2.15 (Local Characterization of Submanifolds). Suppose 𝑀 is a

complex manifold. A subset 𝑆 ⊆ 𝑀 is an embedded complex manifold if and only
if each point of 𝑆 has a neighborhood 𝑈 in 𝑀 such that 𝑈 ∩ 𝑆 is an embedded
complex submanifold of 𝑈 .

► Exercise 2.16. Prove this proposition.

A complex submanifold of codimension 𝑑 is also a real submanifold of codi-

mension 2𝑑. When 𝑑 is positive, this results in the following important property.

Proposition 2.17. Suppose 𝑀 is a connected complex manifold and 𝑆 ⊆ 𝑀

is a closed complex submanifold of positive codimension. Then 𝑀 ∖ 𝑆 is path-
connected.

Proof. Let 𝑛 be the complex dimension of 𝑀 and 𝑘 the complex dimension of 𝑆,

so that 𝑘 < 𝑛. Let 𝑝 and 𝑞 be distinct points in 𝑀 ∖𝑆. Since connected manifolds are
path-connected, there is a continuous path 𝛾 ∶ [0, 1] → 𝑀 such that 𝛾(0) = 𝑝 and
𝛾(1) = 𝑞. For each 𝑡 ∈ [0, 1], we choose a neighborhood 𝑈𝑡 of 𝛾(𝑡) in 𝑀 as follows:
If 𝛾(𝑡) ∉ 𝑆, then because 𝑆 is closed we can choose 𝑈𝑡 such that 𝑈𝑡 ∩ 𝑆 = ∅ and
𝑈𝑡 is path-connected. If 𝛾(𝑡) ∈ 𝑆, let 𝑈𝑡 be the domain of a holomorphic local slice
chart, so 𝑈𝑡 ∩ 𝑆 is a holomorphic 𝑘-slice of 𝑈 . In the latter case, we may shrink 𝑈𝑡
so that it is biholomorphic to a product 𝔹2𝑘 × 𝔹2𝑛−2𝑘 , under a biholomorphism that
sends 𝑈𝑡 ∩ 𝑆 to 𝔹2𝑘 × {0}. Thus 𝑈𝑡 ∖ 𝑆 is biholomorphic to 𝔹2𝑘 × (𝔹2𝑛−2𝑘 ∖ {0}).
Because 2𝑛 − 2𝑘 ≥ 2, the set 𝔹2𝑛−2𝑘 ∖ {0} is path-connected, and thus so is 𝑈𝑡 ∖ 𝑆.
By the Lebesgue number lemma [LeeTM, Lemma 7.18], there are points 0 =
𝑡0 < 𝑡1 < ⋯ < 𝑡𝑚 = 1 such that 𝛾([𝑡𝑖−1 , 𝑡𝑖 ]) is contained in one of the sets 𝑈𝑡
described above for each 𝑖; call this set 𝑈𝑖 . For each 𝑖 = 1, … , 𝑚 − 1, the open
set 𝑈𝑖 ∩ 𝑈𝑖+1 is nonempty because it contains 𝛾(𝑡𝑖 ); and because 𝑆 contains no
nonempty open subset of 𝑀, we can choose a point 𝑥𝑖 ∈ 𝑈𝑖 ∩ 𝑈𝑖+1 ∖ 𝑆. Let 𝑥0 = 𝑝
and 𝑥𝑚 = 𝑞. Then because 𝑈𝑖 ∖ 𝑆 is path connected, for each 𝑖 = 1, … , 𝑚 there is
a path 𝜎𝑖 in 𝑈𝑖 ∖ 𝑆 from 𝑥𝑖−1 to 𝑥𝑖 ; and then the concatenation of these paths is a
path in 𝑀 ∖ 𝑆 from 𝑝 to 𝑞. □
50 2. Complex Submanifolds

Suppose 𝑀 and 𝑁 are complex manifolds and 𝐹 ∶ 𝑀 → 𝑁 is a holomorphic map.

Proposition 2.18 (Restricting Domains or Codomains of Holomorphic Maps).

(a) If 𝑆 ⊆ 𝑀 is an embedded complex submanifold, then 𝐹 |𝑆 ∶ 𝑆 → 𝑁 is

(b) If 𝑇 ⊆ 𝑁 is an embedded complex submanifold such that 𝐹 (𝑀) ⊆ 𝑇 ,

holomorphic.

then 𝐹 is holomorphic as a map from 𝑀 to 𝑇 .

Proof. Part (a) follows because 𝐹 |𝑆 is the composition of the inclusion 𝑆 ↪ 𝑀 fol-
lowed by 𝐹 . For Part (b), it follows from smooth manifold theory that 𝐹 is smooth
as a map into 𝑇 , and then holomorphicity can be checked by using holomorphic
slice coordinates. □

Examples of Complex Submanifolds

The notion of complex submanifolds gives us some tools for defining many
more examples of complex manifolds.
Example 2.19 (Open Submanifolds). If 𝑀 is a complex manifold, every open
subset of 𝑀 is an embedded complex submanifold of codimension 0. //
Example 2.20 (Images of Holomorphic Embeddings). If 𝐹 ∶ 𝑁 → 𝑀 is a holo-
morphic embedding, then the image 𝐹 (𝑁) is an embedded complex submanifold,
directly from the definition. //
Example 2.21 (Graphs of Holomorphic Maps). Suppose 𝑓 ∶ 𝑀 → 𝑁 is a holo-
morphic map. The graph of 𝒇 is the set Γ(𝑓 ) = {(𝑝, 𝑓 (𝑝)) ⊆ 𝑀 × 𝑁 ∶ 𝑝 ∈ 𝑀}.
This is a complex submanifold of 𝑀 × 𝑁, biholomorphic to 𝑀, because the map
from 𝑀 to 𝑀 × 𝑁 given by 𝑝 ↦ (𝑝, 𝑓 (𝑝)) is a holomorphic embedding. //
Example 2.22 (Regular Level Sets). Suppose 𝑀 and 𝑁 are complex manifolds,
Φ ∶ 𝑀 → 𝑁 is a holomorphic map, and 𝑐 ∈ 𝑁 is a regular value of 𝚽 (meaning
that 𝐷Φ(𝑝) is surjective for each 𝑝 ∈ Φ−1 (𝑐)). Then the level set Φ−1 (𝑐) is a com-
plex submanifold of 𝑀 whose codimension is equal to the dimension of 𝑁. The
proof is a simple application of the holomorphic rank theorem and the holomorphic
slice criterion. //
Example 2.23 (Transverse Preimages). Suppose 𝑀 and 𝑁 are complex mani-
folds and 𝑆 ⊆ 𝑁 is a complex submanifold. A holomorphic map 𝐹 ∶ 𝑀 → 𝑁 is
said to be transverse to 𝑺 if for every 𝑥 ∈ 𝐹 −1 (𝑆), the subspaces 𝐷𝐹 (𝑥)(𝑇𝑥 𝑀)
and 𝑇𝐹 (𝑥) 𝑆 together span 𝑇𝐹 (𝑥) 𝑁. If this is the case, then 𝐹 −1 (𝑆) is a complex
submanifold whose codimension in 𝑀 is equal to the codimension of 𝑆 in 𝑁. //

► Exercise 2.24. Prove the above claim: the preimage of a complex submani-
fold under a holomorphic map that is transverse to the submanifold is a complex
submanifold of the same codimension. (See Theorem 6.30 of [LeeSM] for the
smooth version; essentially the same proof goes through in the holomorphic case.)
Complex Submanifolds 51

Example 2.25 (Complex Submanifolds of ℂ𝒏 ). It is important to observe, to begin

with, that there are no compact complex submanifolds of ℂ𝑛 except 0-dimensional
ones. To verify this, note that if 𝑀 ⊆ ℂ𝑛 is a compact complex submanifold, then
each of the coordinate functions 𝑧𝑗 restricts to a global holomorphic function on 𝑀,
which must therefore be constant on each connected component of 𝑀 by Corollary
1.33.

folds arise as complex submanifolds of ℂ𝑛 . Some obvious ones are graphs of holo-
On the other hand, many important examples of noncompact complex mani-

morphic functions from ℂ𝑘 to ℂ𝑛−𝑘 , and level sets of holomorphic submersions

from ℂ𝑛 to ℂ𝑛−𝑘 .

𝑉 ⊆ ℂ𝑛 is called an affine algebraic variety if it is the common zero set of a

A particularly important class of examples is defined as follows: a subset

finite collection 𝑝1 , … , 𝑝𝑘 of holomorphic polynomials. (Some authors require that

varieties, but we do not require this.) A point 𝑧0 ∈ 𝑉 is a regular point of 𝑽 if

a variety be irreducible, meaning that it is not the union of two or more nonempty

it has a neighborhood 𝑈 ⊆ ℂ𝑛 such that 𝑉 ∩ 𝑈 is the common zero set of some

of the polynomials 𝑝𝑖1 , … , 𝑝𝑖𝑟 with the property that the differentials 𝑑𝑝𝑖1 , … , 𝑑𝑝𝑖𝑟
are linearly independent at every point of 𝑉 ∩ 𝑈 ; points that are not regular are
called singular points of 𝑽 . An affine algebraic variety is said to be nonsingular
or smooth if it has no singular points; in this case 𝑉 is a complex submanifold of
ℂ𝑛 by Corollary 2.13. //

Example 2.26 (Complex Lie Subgroups). Suppose 𝐺 is a complex Lie group. An

(embedded) complex submanifold of 𝐺 that is also a subgroup is called a complex
Lie subgroup; it inherits a complex Lie group structure of its own because the group
operations are holomorphic by restriction. Some specific examples:

• The group SL(𝑛, ℂ) of complex matrices with determinant 1 is a com-

plex Lie subgroup of GL(𝑛, ℂ) because it is a level set of the holomorphic
submersion det ∶ GL(𝑛, ℂ) → ℂ.

• O(𝑛, ℂ) denotes the group of 𝑛 × 𝑛 complex matrices 𝐴 satisfying 𝐴𝑇 𝐴 =

Id. It is a regular level set of the holomorphic polynomial function 𝐴 ↦
𝐴𝑇 𝐴, so it is a complex Lie subgroup of GL(𝑛, ℂ).

• SO(𝑛, ℂ) is the subgroup of O(𝑛, ℂ) consisting of matrices of determinant

1; because it is a regular level set of det ∶ O(𝑛, ℂ) → ℂ, it is a complex
Lie subgroup.

It should be noted that the unitary group U(𝑛) ⊆ GL(𝑛, ℂ) is not a complex Lie
group, because it does not have a holomorphic defining function, even locally. (See
Problem 2-1.) //
52 2. Complex Submanifolds

Complex Submanifolds of Projective Spaces

A complex manifold 𝑀 is called a projective manifold if it is biholomorphic to a
closed complex submanifold of ℂℙ𝑛 . Because projective manifolds will be among

standing the structure of submanifolds of ℂℙ𝑛 .

our main objects of study, it is worthwhile to devote a little more time to under-

The most important examples (in fact, thanks to Chow’s theorem discussed

since there are no nonconstant holomorphic functions from ℂℙ𝑛 to ℂ, polynomial

below, the only examples) are algebraic ones, which we will define shortly. But

Suppose 𝑝 ∶ ℂ𝑛+1 → ℂ is a holomorphic polynomial function (that is, a polyno-

or otherwise, we have to be a little more careful about what we mean.

mial function of the variables 𝑤0 , … , 𝑤𝑛 ). It is said to be homogeneous of degree

𝒅 for a nonnegative integer 𝑑 if 𝑝(𝜆𝑤) = 𝜆𝑑 𝑤 for all 𝜆 ∈ ℂ and 𝑤 ∈ ℂ𝑛+1 . This
just means that 𝑝 can be written as a sum of monomial terms each of which has the
same total degree 𝑑. Although such a polynomial does not produce a well-defined
function from ℂℙ𝑛 to ℂ, nonetheless its zero set is a well-defined subset of ℂℙ𝑛 ,
because if 𝑝(𝑤) = 0, then 𝑝(𝜆𝑤) = 0 for all 𝜆 ∈ ℂ ∖ {0}.
A projective algebraic variety is a subset 𝑉 ⊆ ℂℙ𝑛 of the form
𝑉 = {[𝑤] ∈ ℂℙ𝑛 ∶ 𝑝1 (𝑤) = ⋯ = 𝑝𝑘 (𝑤) = 0},
where 𝑝1 , … , 𝑝𝑘 ∶ ℂ𝑛+1 → ℂ are homogeneous holomorphic polynomials of vari-

variety 𝑉̃ ⊆ ℂ𝑛+1 defined by 𝑝1 , … , 𝑝𝑘 has no singular points other than perhaps the
ous degrees. A projective variety is said to be nonsingular or smooth if the affine

variety is actually a complex submanifold of ℂℙ𝑛 . Projective algebraic varieties are

origin. It follows from the next lemma that every nonsingular projective algebraic

the main objects of study in complex algebraic geometry.

Lemma 2.27. Suppose 𝑀 ⊆ ℂℙ𝑛 is a subset and 𝑀 ̃ = 𝜋 −1 (𝑀) ⊆ ℂ𝑛+1 ∖ {0},
where 𝜋 ∶ ℂ ∖ {0} → ℂℙ is the canonical projection. Then 𝑀 is a complex 𝑘-
𝑛+1 𝑛
̃ is a complex (𝑘 + 1)-dimensional submanifold
submanifold of ℂℙ𝑛 if and only if 𝑀
of ℂ 𝑛+1
∖ {0}.

Proof. Problem 2-2. □

Lemma 2.28. Every projective algebraic variety in ℂℙ𝑛 is compact and therefore
closed in ℂℙ𝑛 .

Proof. Let 𝑉 ⊆ ℂℙ𝑛 be the variety determined by homogeneous polynomials

𝑝1 , … , 𝑝𝑘 ∶ ℂ𝑛+1 → ℂ, and let 𝑉̃ = 𝜋 −1 (𝑉 ) ⊆ ℂ𝑛+1 ∖ {0}, where 𝜋 ∶ ℂ𝑛+1 ∖ {0}
is the canonical projection. By homogeneity, every point [𝑤] ∈ 𝑉 has a preimage
𝑤′ = 𝑤/|𝑤| that lies in 𝑉̃ ∩ 𝕊2𝑛+1 . Thus 𝑉 is the image under 𝜋 of the compact set
𝑉̃ ∩ 𝕊2𝑛+1 , so it is compact. Since ℂℙ𝑛 is a compact Hausdorff space, all compact
subsets of it are closed. □
Complex Submanifolds of Projective Spaces 53

The algebraic submanifolds of projective space might seem like a restrictive

special class, but in fact they are quite general, thanks to the following remarkable
theorem proved in 1949 by Wei-Liang Chow [Cho49].

Theorem 2.29 (Chow’s Theorem). Every closed complex submanifold of ℂℙ𝑛 is

algebraic.

Later, we will give a proof of Chow’s theorem for the special case of hypersur-
faces (see Cor. 9.53). We will neither prove nor use the general case, but you can

𝑉 ⊆ ℂℙ𝑛 is a closed analytic variety, meaning that each point of 𝑉 has a neigh-
find a proof in [GH94, p. 167]. (Actually, Chow proved something stronger—if

borhood 𝑈 such that 𝑉 ∩ 𝑈 is the common zero set of finitely many holomorphic
functions defined on 𝑈 , then 𝑉 is actually a projective algebraic variety. In 1956,
Jean-Pierre Serre introduced a vast generalization of Chow’s theorem in a famous
paper called “Géométrie algébrique et géométrie analytique” [Ser55b]: roughly
speaking, it showed that virtually any geometric structure that can be defined holo-
morphically in complex projective space can actually be defined algebraically. That
general fact is now known as the GAGA principle, after the French title of Serre’s
paper.)
We note in passing that in algebraic geometry, there is another topology that

on ℂℙ𝑛 is defined by declaring a set to be open if and only if its complement is

is usually used for affine or projective algebraic varieties. The Zariski topology

an algebraic variety; the Zariski topology on ℂ𝑛 is defined similarly. The Zariski

topology on a projective or affine algebraic variety is then defined as the induced
subspace topology. From an algebraic point of view, this topology is advantageous
because algebraic properties of subsets are directly reflected in topological prop-
erties, and because the topology makes sense for algebraic varieties defined over

of the usual manifold topologies on ℂ𝑛 and ℂℙ𝑛 , most notably the fact that is never
arbitrary fields. But the Zariski topology has properties quite different from those

Hausdorff except in the 0-dimensional case. In this book, we will use only the
standard topologies on ℂ𝑛 and ℂℙ𝑛 and their subsets.
Algebraic varieties in ℂℙ1 are easy to describe.

Proposition 2.30. A subset 𝑉 ⊆ ℂℙ1 is an algebraic variety if and only if it is

either finite or all of ℂℙ1 .

Proof. Suppose 𝑉 = {𝑎1 , … , 𝑎𝑘 } is a finite subset of ℂℙ1 . If 𝑉 = ∅, then it is the

variety determined by the polynomial 𝑓 ≡ 1. Otherwise, for each 𝑗, we can write
𝑎𝑗 = [𝑎0𝑗 , 𝑎1𝑗 ] for some point (𝑎0𝑗 , 𝑎1𝑗 ) ∈ ℂ2 , so {𝑎𝑗 } is the variety determined by the
polynomial 𝑓𝑗 (𝑤0 , 𝑤1 ) = 𝑎1𝑗 𝑤0 − 𝑎0𝑗 𝑤1 . It follows that 𝑉 is the variety determined
by the product 𝑓1 ⋅ ⋯ ⋅ 𝑓𝑘 . On the other hand, ℂℙ1 itself is the variety determined
by the zero polynomial.
54 2. Complex Submanifolds

Conversely, if 𝑉 is the variety determined by the homogeneous polynomials

𝑓1 , … , 𝑓𝑘 , and 𝑉 is not equal to ℂℙ1 , then at least one of the polynomials, say 𝑓𝑗 ,
is not identically zero. The points of 𝑉 in ℂℙ1 ∖ {[0, 1]} can all be written in the
form [1, 𝑧] where 𝑓𝑗 (1, 𝑧) = 0. The polynomial function 𝑧 ↦ 𝑓𝑗 (1, 𝑧) has only
finitely many zeros by the fundamental theorem of algebra. □

Our definition of a projective algebraic variety 𝑉 ⊆ ℂℙ𝑛 of codimension 𝑘

allows for the possibility that more than 𝑘 homogeneous polynomials are required
to define it. It turns out that for nonsingular projective hypersurfaces, a single poly-
nomial always suffices. We will not have the tools to prove this until Chapter 9, but
we state it here for convenience because it will make some results easier to state.
For the proof, see Corollary 9.52.

If 𝑉 ⊆ ℂℙ𝑛 is a codimension-1 nonsingular algebraic variety, then 𝑉 is the variety

Proposition 2.31 (Algebraic Hypersurface Is Cut Out by a Single Polynomial).

defined by a single homogeneous polynomial 𝑝 ∶ ℂ𝑛+1 → ℂ.

To understand submanifolds of ℂℙ𝑛 more deeply, we will use the fact that ℂℙ𝑛
has a lot of symmetry. Because GL(𝑛 + 1, ℂ) acts transitively on 1-dimensional
subspaces of ℂ𝑛+1 , it follows that the group of projective transformations acts tran-
sitively on ℂℙ𝑛 . Two subsets 𝑆1 , 𝑆2 ⊆ ℂℙ𝑛 are said to be projectively equivalent
if there is a projective transformation taking 𝑆1 to 𝑆2 .
Let us explore some particular complex submanifolds of ℂℙ𝑛 . First of all, if
𝑝1 , … , 𝑝𝑛−𝑘 are linearly independent holomorphic linear functions on ℂ𝑛+1 , the
variety 𝑉 they determine is a 𝑘-dimensional complex submanifold of ℂℙ𝑛 called a
projective linear subspace. If we choose a basis 𝑏0 , … , 𝑏𝑘 for the common kernel
of 𝑝1 , … , 𝑝𝑛−𝑘 in ℂ𝑛+1 , then the map [𝑤0 , … , 𝑤𝑘 ] ↦ [𝑤0 𝑏0 + ⋯ + 𝑤𝑘 𝑏𝑘 ] is a holo-
morphic embedding of ℂℙ𝑘 into ℂℙ𝑛 , whose image is exactly the variety 𝑉 . Thus
every 𝑘-dimensional projective linear subspace of ℂℙ𝑛 is biholomorphic to ℂℙ𝑘 .
In particular, a 1-dimensional projective linear subspace is called a projective line;
a 2-dimensional one is called a projective plane; and an (𝑛 − 1)-dimensional one
in ℂℙ𝑛 is called a projective hyperplane. Since any two (𝑘 + 1)-dimensional sub-
spaces of ℂ𝑛+1 are related by a holomorphic linear isomorphism, it follows that any
two 𝑘-dimensional projective linear subspaces of ℂℙ𝑛 are projectively equivalent.

Example 2.32 (Dual Projective Spaces). For any positive integer 𝑛, the set of
projective hyperplanes in ℂℙ𝑛 is called the dual projective space to ℂℙ𝑛 , and de-
noted by (ℂℙ𝑛 )∗ . Since each projective hyperplane is the variety determined by
a nonzero complex-linear functional 𝑓 ∈ (ℂ𝑛+1 )∗ , and two such functionals de-

(ℂℙ𝑛 )∗ is canonically identified with the projectivization ℙ((ℂ𝑛+1 )∗ ), and there-

termine the same variety if and only if they differ by a nonzero complex multiple,

fore (ℂℙ𝑛 )∗ is biholomorphic to ℂℙ𝑛 . One such biholomorphism is obtained by

sending [𝑎0 , … , 𝑎𝑛 ] ∈ ℂℙ𝑛 to the variety determined by the linear functional
Complex Submanifolds of Projective Spaces 55

𝑓𝑎 (𝑤) = 𝑎0 𝑤0 + ⋯ + 𝑎𝑛 𝑤𝑛 . (But note that this biholomorphism is heavily de-

pendent on working with the standard basis of ℂ𝑛+1 ; another choice of basis will
result in a different isomorphism between ℂℙ𝑛 and (ℂℙ𝑛 )∗ .) //
Now consider the map 𝐼 ∶ ℂ𝑛 → ℂℙ𝑛 given by 𝐼(𝑧1 , … , 𝑧𝑛 ) = [1, 𝑧1 , … , 𝑧𝑛 ].
It is the inverse of one of the holomorphic coordinate charts we defined in Example

complement of the projective hyperplane Π defined by 𝑧0 = 0, so as a set, ℂℙ𝑛 is

1.9, and thus it is a biholomorphism onto its image. That image is exactly the

the disjoint union of 𝐼(ℂ𝑛 ) and Π. We can think of ℂℙ𝑛 as the union of a copy of
ℂ𝑛 together with the (projective) “hyperplane at infinity.”
By following the embedding 𝐼 with a suitable projective transformation, we can
obtain a holomorphic embedding of ℂ𝑛 into ℂℙ𝑛 whose image is the complement

embedding of ℂ𝒏 ; the embedding 𝐼 described above is called the standard affine

of any arbitrary projective hyperplane. Any such embedding is called an affine

embedding. Thus any nontrivial complex-linear functional 𝑝 on ℂ𝑛+1 determines a

decomposition of ℂℙ𝑛 into a projective hyperplane (determined by the kernel of 𝑝)
and its complement, which is a dense, open subspace that is the image of an affine

Proposition 1.57 showed that the holomorphic tangent space to a point of ℂ𝑛

embedding.

is canonically identified with ℂ𝑛 itself, from which it follows that the holomorphic
tangent space to a complex submanifold of ℂ𝑛 can be identified with a linear sub-
space of ℂ𝑛 . The next proposition gives a projective version of this identification.
Proposition 2.33 (The Projective Tangent Space). Suppose 𝑀 ⊆ ℂℙ𝑛 is a 𝑘-
dimensional complex submanifold. For each 𝑝 ∈ 𝑀, there is a unique 𝑘-dimen-
sional projective linear subspace Π ⊆ ℂℙ𝑛 containing 𝑝 with the property that
𝑇𝑝′ 𝑀 = 𝑇𝑝′ Π. It is called the projective tangent space to 𝑴 at 𝒑.

Proof. First we prove existence. Let 𝑀 ̃ = 𝜋 −1 (𝑀) ⊆ ℂ𝑛+1 ∖ {0}. Lemma 2.27
shows that 𝑀 ̃ is a (𝑘 + 1)-dimensional complex submanifold. Given 𝑝 ∈ 𝑀, let
𝑝 ̃ be any point in 𝜋 −1 (𝑝). Since 𝜋 maps 𝑀 ̃ to 𝑀, the linear map 𝐷′ 𝜋(𝑝)̃ maps
′ ̃
𝑇𝑝 ̃ 𝑀 into 𝑇𝑝 𝑀. The kernel of 𝐷 𝜋(𝑝)̃ is spanned by the vector 𝑝𝑗̃ 𝜕/𝜕𝑤𝑗 |𝑝 .̃ This
′ ′

̃ so when 𝐷′ 𝜋(𝑝)̃ is restricted to the (𝑘 + 1)-dimensional

span is contained in 𝑇𝑝′̃ 𝑀,
subspace 𝑇𝑝′̃ 𝑀̃ ⊆ 𝑇𝑝′̃ ℂ𝑛+1 , its rank is exactly 𝑘. Thus it is surjective onto 𝑇𝑝′ 𝑀 for
dimensional reasons: 𝑇𝑝′ 𝑀 = 𝐷′ 𝜋(𝑝)̃ (𝑇𝑝′̃ 𝑀 ̃ ).
Under the canonical identification of 𝑇𝑝′̃ ℂ𝑛+1 with ℂ𝑛+1 itself (see Prop. 1.57),
we may consider 𝑇𝑝′̃ 𝑀 ̃ as a (𝑘+1)-dimensional linear subspace Π ̃ ⊆ ℂ𝑛+1 . Because
′ ̃ ̃ The image of Π
𝑝 ̃ 𝜕/𝜕𝑤 |𝑝 ̃ ∈ 𝑇𝑝 ̃ 𝑀, it follows that 𝑝 ̃ ∈ Π.
𝑗 𝑗 ̃ in ℂℙ is a 𝑘-dimensional
𝑛

projective linear subspace Π containing 𝑝, and the argument above with Π ̃ in place
of 𝑀 ̃ shows that
̃ ) = 𝐷′ 𝜋(𝑝)̃ (𝑇𝑝′̃ 𝑀
𝑇𝑝′ Π = 𝐷′ 𝜋(𝑝)̃ (𝑇𝑝′̃ Π ̃ ) = 𝑇𝑝′ 𝑀.

Thus Π is the projective subspace we seek.

56 2. Complex Submanifolds

̃ ′ = 𝜋 −1 (Π′ ), then 𝑇𝑝′̃ Π

If Π′ is another such subspace and Π ̃ ′ contains the vector
𝑝 ̃ 𝜕/𝜕𝑧 |𝑝 ̃ and is mapped onto 𝑇𝑝 𝑀 by 𝐷 𝜋(𝑝).
𝑗 𝑗 ′ ′
̃ It follows that 𝑇𝑝′̃ Π ̃ ′ = 𝑇𝑝′̃ Π,
̃ from
which it follows that Π ̃ =Π
′ ̃ and therefore Π = Π.
′
□

The case of ℂℙ1 is especially notable. In this case, a projective hyperplane

is just a single point, so ℂℙ1 is the union of a copy of ℂ (which we may identify
with the set of points of the form [1, 𝑧] under the standard affine embedding) to-
gether with the “point at infinity” (the point [0, 1], which we denote by ∞ in this

ℂ, also called the Riemann sphere. In fact, Problem 2-4 shows that ℂℙ1 is diffeo-
context). Topologically, it is homeomorphic to the one-point compactification of

morphic to 𝕊2 .
Thanks to the result of Problem 2-9, every automorphism of ℂℙ1 is a map of
the form 𝑚([𝑤, 𝑧]) = [𝑐𝑧 + 𝑑𝑤, 𝑎𝑧 + 𝑏𝑤] for some complex numbers 𝑎, 𝑏, 𝑐, 𝑑 with
𝑎𝑑 − 𝑏𝑐 ≠ 0 (to ensure that the corresponding linear map on ℂ2 is invertible).
Assuming 𝑐 ≠ 0, we can write this map in affine coordinates (defined by 𝑤 = 1) as
𝑎𝑧 + 𝑏
⎧
⎪ 𝑐𝑧 + 𝑑 , 𝑧 ≠ ∞, −𝑑/𝑐,
⎪
𝑚(𝑧) = ⎨ 𝑎 , 𝑧 = ∞,
⎪𝑐
⎪∞, 𝑧 = −𝑑/𝑐.
⎩
In the remaining case 𝑐 = 0, it just maps 𝑧 ∈ ℂ to (𝑎𝑧 + 𝑏)/𝑑 and ∞ to ∞. Any
such automorphism is called a Möbius transformation.
The higher-dimensional projective spaces are not diffeomorphic to spheres. But
we do have the following.
Proposition 2.34. For every 𝑛 ≥ 1, ℂℙ𝑛 is simply connected.

Proof. We will prove this by induction on 𝑛. For 𝑛 = 1, it follows from the fact that
ℂℙ1 is diffeomorphic to 𝕊2 , which is simply connected. So suppose 𝑛 ≥ 1 and we
have shown that ℂℙ𝑛 is simply connected. We can write ℂℙ𝑛+1 = 𝑈 ∪ 𝑉 , where
𝑈 = {[𝑤0 , … , 𝑤𝑛+1 ] ∈ ℂℙ𝑛+1 ∶ 𝑤0 ≠ 0},
𝑉 = ℂℙ𝑛+1 ∖ {[1, 0, … , 0]}.
Then 𝑈 is the image of the standard affine embedding and thus biholomorphic to
ℂ𝑛+1 , so it is simply connected. We will show that 𝑉 is homotopy equivalent to
ℂℙ𝑛 . Consider the map 𝐻 ∶ 𝑉 × [0, 1] → 𝑉 given by
𝐻([𝑤0 , … , 𝑤𝑛+1 ], 𝑡) = [𝑡𝑤0 , 𝑤1 , … , 𝑤𝑛+1 ].
If we let Π denote the projective hyperplane defined by 𝑤0 = 0, then Π is biholo-
morphic to ℂℙ𝑛 , and 𝐻 is a strong deformation retraction of 𝑉 onto Π. It follows
from the induction hypothesis that 𝑉 is also simply connected, so the Seifert–Van
Kampen theorem (specifically, [LeeTM, Cor. 10.5]) shows that ℂℙ𝑛+1 is simply
connected. □
Complex Submanifolds of Projective Spaces 57

Next we explore some examples of algebraic submanifolds of ℂℙ𝑛 of degree

higher than 1. Suppose 𝑆 ⊆ ℂℙ𝑛 is a projective algebraic variety defined by a
single homogeneous polynomial of degree 𝑑; in this case we say 𝑑 is the degree of
the variety. Projective varieties cut out by single polynomials of degree 2, 3, 4, and
5 are called projective quadrics, cubics, quartics, and quintics, respectively.
To study projective quadrics, we begin with some general considerations about
homogeneous quadratic polynomials.
Lemma 2.35. Suppose 𝑝 is a homogeneous quadratic holomorphic polynomial
function on an 𝑛-dimensional complex vector space 𝑉 . It is always possible to
find a basis for 𝑉 in which 𝑝 has the coordinate representation
(2.1) 𝑝(𝑤1 , … , 𝑤𝑛 ) = (𝑤1 )2 + ⋯ + (𝑤𝑟 )2
for some 0 ≤ 𝑟 ≤ 𝑛. The integer 𝑟, called the rank of 𝒑, is independent of the choice
of basis.

gorithm. Begin by choosing a basis for 𝑉 and writing 𝑝 in the form 𝑝(𝑧) =
Proof. The proof is essentially a complex-linear version of the Gram-Schmidt al-

∑𝑗,𝑘 𝑝𝑖𝑗 𝑧𝑖 𝑧𝑗 . We can assume that the matrix 𝑝𝑖𝑗 is symmetric—if not, just replace
𝑝𝑖𝑗 with 21 (𝑝𝑖𝑗 + 𝑝𝑗𝑖 ), which does not change the values of the polynomial 𝑝. Define
𝐵 ∶ 𝑉 × 𝑉 → ℂ by 𝐵(𝑧, 𝑤) = ∑𝑗,𝑘 𝑝𝑖𝑗 𝑧𝑖 𝑤𝑗 , so that 𝐵 is a symmetric complex-
bilinear form satisfying 𝐵(𝑧, 𝑧) = 𝑝(𝑧). Note that 𝐵 can also be written in the form
𝐵(𝑧, 𝑤) = 41 (𝑝(𝑧 + 𝑤) − 𝑝(𝑧 − 𝑤)), so it is independent of the choice of basis.
We will prove by induction on 𝑛 = dim 𝑉 that 𝑉 has a basis in which 𝐵 has the
form 𝐵(𝑧, 𝑤) = 𝑧1 𝑤1 + ⋯ + 𝑧𝑟 𝑤𝑟 for some 𝑟, which implies the result. For 𝑛 = 0,

Assume the result is true for spaces of dimension 𝑛 − 1 and suppose 𝑉 has
there is nothing to prove.

dimension 𝑛. If 𝐵 ≡ 0, any basis will do. Otherwise, there is some 𝑏1 ∈ 𝑉 such

that 𝐵(𝑏1 , 𝑏1 ) ≠ 0. After multiplying 𝑏1 by an appropriate complex scalar, we can
arrange that 𝐵(𝑏1 , 𝑏1 ) = 1. Let 𝑆 ⊆ 𝑉 be the set {𝑧 ∈ 𝑉 ∶ 𝐵(𝑧, 𝑏1 ) = 0}. Because
𝑧 ↦ 𝐵(𝑧, 𝑏1 ) is a nonzero linear functional, 𝑆 is a complex-linear subspace of
dimension 𝑛 − 1. The induction hypothesis shows that 𝑆 has a basis (𝑏2 , … , 𝑏𝑛 ) in
which the restriction of 𝐵 to 𝑆 has the coordinate formula 𝐵(𝑧, 𝑤) = 𝑧2 𝑤2 + ⋯ +
𝑧𝑟 𝑤𝑟 , and then a simple computation shows that the basis (𝑏1 , … , 𝑏𝑛 ) for 𝑉 has the

To prove that the rank is independent of the choice of basis, just note that 𝑛 − 𝑟
desired property.

is equal to the largest dimension of a complex-linear subspace 𝑊 ⊆ 𝑉 such that

𝐵(𝑤, 𝑧) = 0 for all 𝑤 ∈ 𝑊 and 𝑧 ∈ 𝑉 . □

Proposition 2.36. Suppose 𝑝 is a holomorphic homogeneous quadratic polynomial

on ℂ𝑛+1 . The variety 𝑉 ⊆ ℂℙ𝑛 defined by 𝑝 is nonsingular if and only if the rank
of 𝑝 is equal to 𝑛 + 1.
58 2. Complex Submanifolds

Proof. After a complex-linear change of basis, we may assume that 𝑝 has the form
𝑝(𝑤) = (𝑤0 )2 + ⋯ + (𝑤𝑟−1 )2 , where 𝑟 is the rank of 𝑝. If 𝑟 = 𝑛 + 1, then we can
write the differential of 𝑝 on ℂ𝑛+1 as

𝑑𝑝 = 2(𝑤0 𝑑𝑤0 + ⋯ + 𝑤𝑛 𝑑𝑤𝑛 ).

This never vanishes except at the origin, showing that 𝑉 is nonsingular.

Conversely, suppose 𝑟 < 𝑛 + 1. In the affine coordinate chart (𝑧1 , … , 𝑧𝑛 ) ↔
[𝑧 , … , 𝑧𝑛 , 1], 𝑉 is the set of points where (𝑧1 )2 + ⋯ + (𝑧𝑟 )2 = 0. All of the
1

following smooth curves starting at the origin lie in 𝑉 :

𝑡 ↦ (𝑡, 0, … , 0, 𝑖𝑡, 0, … , 0), 𝑡 ↦ (𝑖𝑡, 0, … , 0, 𝑡, 0, … , 0), 𝑗 = 2, … , 𝑛

𝑡 ↦ (𝑡, −𝑖𝑡, 0, … , 0), 𝑡 ↦ (𝑖𝑡, −𝑡, 0, … , 0)

(where the nonzero entries in the first line are in positions 1 and 𝑗). The initial
velocity vectors of these curves span the entire (real) tangent space to ℂ𝑛 at the ori-
gin. Since elsewhere 𝑉 is a complex codimension-1 submanifold and thus a smooth
submanifold of real codimension 2, this shows that 𝑉 is not a smooth submanifold
in a neighborhood of the origin in this affine chart. □

Proposition 2.37. All nonsingular projective quadrics in ℂℙ𝑛 are projectively

equivalent.

Proof. If 𝑉 ⊆ ℂℙ𝑛 is a nonsingular projective quadric defined by a quadratic poly-

nomial 𝑝, Proposition 2.36 shows that 𝑝 has rank 𝑛 + 1, and then Lemma 2.35 shows
that a change of basis in ℂ𝑛+1 (which induces a projective transformation on ℂℙ𝑛 ),
puts 𝑝 into the standard form (𝑤0 )2 + ⋯ + (𝑤𝑛 )2 . □

Problems 2-5 and 2-6 show that nonsingular quadrics in ℂℙ2 are biholomorphic
to ℂℙ1 , and those in ℂℙ3 are biholomorphic to ℂℙ1 ×ℂℙ1 . The higher-dimensional
ones are diffeomorphic to certain Grassmann manifolds; see Problem 2-7.
Now let us look at some more general projective manifolds. In general, one
can get an idea of the structure of an algebraic variety by expressing it in affine

coordinates to 1, which produces an affine algebraic variety that typically includes

coordinates; computationally, this just amounts to setting one of the homogeneous

an open dense subset of the corresponding projective variety. That is what we did

Conversely, given an affine algebraic hypersurface 𝑉 ⊆ ℂ𝑛 determined by a

in the proof of Proposition 2.36, for example.

holomorphic polynomial 𝑝, we can complete it to a projective variety in the fol-

lowing way. Suppose 𝑝 ∶ ℂ𝑛 → ℂ is a holomorphic polynomial of degree 𝑚. By
𝑑=0 𝑝(𝑑) , where 𝑝(𝑑) is
separating the monomial terms by degree, we can write 𝑝 = ∑𝑚
a homogenenous polynomial of degree 𝑑. Then define a homogeneous polynomial
Complex Submanifolds of Projective Spaces 59

𝑝 ̃ ∶ ℂ𝑛+1 → ℂ, called the homogenization of 𝒑, by

𝑚
𝑝(𝑤
̃ 0 , … , 𝑤𝑛 ) = (𝑤0 )𝑚−𝑑 𝑝(𝑑) (𝑤1 , … , 𝑤𝑛 ).
∑
𝑑=0

This homogeneous polynomial defines a projective hypersurface 𝑉 ⊆ ℂℙ𝑛 , called

the projective completion of 𝑽 . The intersection of 𝑉 with the standard embedded
copy of ℂ𝑛 is the original affine algebraic variety. Even if the original variety 𝑉
was smooth, though, 𝑉 might have singularities on the hyperplane at infinity, which
must be checked separately. (Projective completions can also be defined for affine
algebraic varieties of higher codimension, but this requires more delicate algebraic
considerations.)
Example 2.38. Consider the affine cubic 𝑉 = {(𝑧, 𝑤) ∶ 𝑤 = 𝑧3 } ⊆ ℂ2 . It is the

Its projective completion is the variety 𝑉 = {[𝜁, 𝑧, 𝑤] ∈ ℂℙ2 ∶ 𝜁 2 𝑤 − 𝑧3 = 0}.

graph of a holomorphic function and thus a nonsingular affine algebraic variety.

The projective hyperplane at infinity is the set of points of the form [0, 𝑧, 𝑤], and 𝑉
contains one point on that hyperplane, namely [0, 0, 1]. To analyze the structure of
the variety near that point, we can switch to affine coordinates (𝛼, 𝛽) ↔ [𝛼, 𝛽, 1]. In
these coordinates 𝑉 has the equation 𝛼 2 − 𝛽 3 = 0. This polynomial has a singular
point at the origin, which can be verified as follows: suppose 𝛾(𝑡) = (𝑓 (𝑡), 𝑔(𝑡)) is
a smooth curve lying in 𝑉 with 𝛾(0) = 0. Then 𝑓 (𝑡)2 ≡ 𝑔(𝑡)3 . Taking two deriva-
tives and setting 𝑡 = 0 shows that 𝑓 ′ (0) = 0, and then taking another derivative
shows that 𝑔 ′ (0) = 0. Thus every tangent vector to 𝑉 at the origin is zero. If 𝑉

dimension 2.
were a nonsingular variety, its tangent space at the origin would have to have real
//
Example 2.39 (Fermat Hypersurfaces). The algebraic hypersurface in ℂℙ𝑛 de-
fined in homogeneous coordinates by the equation
(2.2) (𝑤0 )𝑑 + ⋯ + (𝑤𝑛 )𝑑 = 0
for an integer 𝑑 > 0 is called a Fermat hypersurface of degree 𝑑. Problem 2-8
shows that each such hypersurface is nonsingular. In the special case 𝑛 = 2, it is
called a Fermat curve. (The name reflects the fact that the equation (𝑤0 )𝑑 +(𝑤1 )𝑑 +
(𝑤2 )𝑑 = 0 is projectively equivalent under the change of variables 𝑤0 ↦ 𝑖𝑤0 to the
equation (𝑤1 )𝑑 + (𝑤2 )𝑑 = (𝑤0 )𝑑 , which is the basis of Fermat’s last theorem.) //

surfaces in ℂℙ𝑛 . When 𝑛 ≥ 2, it is easy to see that two projective hyperplanes

For future use, we record the following important property of algebraic hyper-

in ℂℙ𝑛 must intersect, because two linear codimension-1 subspaces in ℂ𝑛+1 must
have a nontrivial intersection by linear algebra. The next lemma generalizes this to
(possibly singular) algebraic hypersurfaces of any degree.
Lemma 2.40. Let 𝑛 ≥ 2. If 𝑉 , 𝑊 ⊆ ℂℙ𝑛 are algebraic varieties in ℂℙ𝑛 , each
defined by a single homogeneous polynomial, then 𝑉 ∩ 𝑊 ≠ ∅.
60 2. Complex Submanifolds

Proof. We will prove this by induction on 𝑛. We begin with the case 𝑛 = 2. Let
[𝑤0 , 𝑤1 , 𝑤2 ] denote the homogeneous coordinates on ℂℙ2 . Suppose 𝑉 , 𝑊 ⊆ ℂℙ2
are the varieties determined by homogeneous polynomials 𝑃 , 𝑄 ∶ ℂ3 → ℂ of de-
grees 𝑑 and 𝑒, respectively. After a change of basis if necessary, we may assume that
𝑃 (0, 0, 1) and 𝑄(0, 0, 1) are both nonzero; this means that the coefficient of (𝑤2 )𝑑
in 𝑃 and that of (𝑤2 )𝑒 in 𝑄 are nonzero. After dividing by suitable constants, we
may assume that both these coefficients are equal to 1.
Define polynomials 𝑝, 𝑞 ∶ ℂ2 → ℂ by
𝑝(𝑥, 𝑦) = 𝑃 (1, 𝑥, 𝑦), 𝑞(𝑥, 𝑦) = 𝑄(1, 𝑥, 𝑦).
By collecting terms in 𝑦, we can write
𝑝(𝑥, 𝑦) = 𝑦𝑑 + 𝑎1 (𝑥)𝑦𝑑−1 + ⋯ + 𝑎𝑑−1 (𝑥)𝑦 + 𝑎𝑑 (𝑥),
for some one-variable polynomials 𝑎1 , … , 𝑎𝑑 . By the fundamental theorem of al-
gebra, for each 𝑥 ∈ ℂ the polynomial 𝑝(𝑥, ⋅) has 𝑑 roots 𝜆1 (𝑥), … , 𝜆𝑑 (𝑥) (listed in
some order, possibly with some roots repeated), and we can write
𝑝(𝑥, 𝑦) = (𝑦 − 𝜆1 (𝑥)) ⋅ ⋯ ⋅ (𝑦 − 𝜆𝑑 (𝑥)).
Let 𝑟 ∶ ℂ → ℂ be the function
𝑟(𝑥) = 𝑞(𝑥, 𝜆1 (𝑥)) ⋅ ⋯ ⋅ 𝑞(𝑥, 𝜆𝑑 (𝑥)),
called the resultant of 𝒑 and 𝒒. Although the 𝜆𝑗 ’s are not polynomial functions
of 𝑥, it turns out that 𝑟 is a polynomial in 𝑥. To see why, note that the function
𝑅 ∶ ℂ𝑑+1 → ℂ given by
𝑅(𝑥, 𝑘1 , … , 𝑘𝑑 ) = 𝑞(𝑥, 𝑘1 ) ⋅ ⋯ ⋅ 𝑞(𝑥, 𝑘𝑑 )
is a symmetric polynomial in 𝑘 = (𝑘1 , … , 𝑘𝑑 ) with coefficients given by polyno-
mials in 𝑥, meaning it is unchanged by applying any permutation to the coordinates
of 𝑘. A basic theorem in algebra called the fundamental theorem on symmetric

elementary symmetric polynomials 𝜎1 (𝑘), … 𝜎𝑑 (𝑘), where 𝜎𝑗 (𝑘) is defined as the

polynomials says that any such polynomial can be expressed as a polynomial in the

coefficient of 𝑦𝑑−𝑗 in the expansion

(𝑦 − 𝑘1 ) ⋅ ⋯ ⋅ (𝑦 − 𝑘𝑑 ) = 𝑦𝑑 + 𝜎1 (𝑘)𝑦𝑑−1 + ⋯ + 𝜎𝑑−1 (𝑘)𝑦 + 𝜎𝑑 (𝑘),
so that 𝜎𝑗 (𝑘) is the sum of all possible 𝑗-fold products of distinct components of
𝑘. (See [Lan02, Thm. 6.1 on p. 191] for a proof of this theorem.) Since the coef-
ficients 𝑎𝑗 (𝑥) of 𝑝 are exactly the elementary symmetric polynomials in the roots
𝜆1 (𝑥), … , 𝜆𝑑 (𝑥), it follows that 𝑟(𝑥) = 𝑅(𝑥, 𝜆1 (𝑥), … , 𝜆𝑑 (𝑥)) is a polynomial func-
tion of (𝑥, 𝑎1 (𝑥), … , 𝑎𝑑 (𝑥)), and therefore a polynomial function of 𝑥. By the fun-
damental theorem of algebra, it has a root 𝑥0 ∈ ℂ. This implies that one of the
factors 𝑞(𝑥0 , 𝜆𝑗 (𝑥0 )) in the definition of 𝑟 must vanish. Since 𝑝(𝑥0 , 𝜆𝑗 (𝑥0 )) = 0
by definition, it follows that the point [𝑥0 , 𝜆𝑗 (𝑥0 ), 1] lies on both 𝑉 and 𝑊 . This
completes the 𝑛 = 2 case.
Complex Submanifolds of Projective Spaces 61

Now let 𝑛 ≥ 2 and assume the result is true for ℂℙ𝑛 , and let 𝑉 , 𝑊 be algebraic
varieties in ℂℙ𝑛+1 defined by homogeneous polynomials 𝑃̃, 𝑄. ̃ Let Π ⊆ ℂℙ𝑛+1
be the projective hyperplane determined by 𝑤𝑛+1 = 0; it is biholomorphic to ℂℙ𝑛 .
After a projective transformation, we may assume that Π is not contained in either
𝑉 or 𝑊 . Define 𝑃 , 𝑄 ∶ ℂ𝑛+1 → ℂ by

𝑃 (𝑤0 , … , 𝑤𝑛 ) = 𝑃̃(𝑤0 , … , 𝑤𝑛 , 0), ̃ 0 , … , 𝑤𝑛 , 0).

𝑄(𝑤0 , … , 𝑤𝑛 ) = 𝑄(𝑤

Then 𝑃 and 𝑄 are nontrivial homogeneous polynomials, and the varieties in Π ≈

ℂℙ𝑛 they determine are 𝑉 ′ = 𝑉 ∩Π and 𝑊 ′ = 𝑊 ∩Π. By the inductive hypothesis,
𝑉 ′ ∩ 𝑊 ′ is nonempty, which shows that 𝑉 and 𝑊 intersect. □

With more work, the argument above in the case of ℂℙ2 can be adapted to
prove the stronger result known as Bézout’s theorem: if 𝑉 , 𝑊 ⊆ ℂℙ2 are distinct
projective algebraic curves defined by irreducible polynomials of degrees 𝑑 and 𝑒,
respectively, then 𝑉 ∩ 𝑊 contains exactly 𝑑𝑒 points counted with multiplicity. See

Although nonsingular projective quadrics in ℂℙ𝑛 are all projectively equiva-

[Har77, Cor. 7.8] for a proof.

gree: as long as 𝑛 ≥ 1 and 𝑑 ≥ 3, there are nonsingular projective hypersurfaces of

lent, the analogous statement is not true for projective hypersurfaces of higher de-

degree 𝑑 in ℂℙ𝑛 that are not biholomorphic to each other. However, it is a remark-
able fact that as smooth manifolds, they are all diffeomorphic, as Theorem 2.43 will
show below. We will not use this result anywhere in the book; but it provides an in-
teresting insight into the nature of projective hypersurfaces. The proof will depend
on two preliminary results: one from differential geometry that we will prove, and
one from commutative algebra that we will accept without proof.
First, the result from differential geometry.
Lemma 2.41. Suppose 𝑀 and 𝑁 are smooth manifolds and 𝜋 ∶ 𝑀 → 𝑁 is a
proper smooth submersion. If 𝑁 is connected, then all fibers of 𝜋 are diffeomorphic
to each other.

Proof. For each smooth vector field 𝑌 on 𝑁, we will show that there exists a lift
of 𝒀 , that is, a smooth vector field 𝑋 on 𝑀 that is 𝜋-related to 𝑌 , meaning that
𝐷𝜋(𝑞)(𝑋𝑞 ) = 𝑌𝜋(𝑞) for each 𝑞 ∈ 𝑀. To see this, note first that the rank theorem
shows that for each 𝑞 ∈ 𝑀 there are smooth coordinate charts (𝑈 , (𝑢1 , … , 𝑢𝑚 )) for
𝑀 containing 𝑞 and (𝑉 , (𝑣1 , … , 𝑣𝑛 )) for 𝑁 containing 𝜋(𝑞) in which 𝜋 has the local
expression 𝜋(𝑢1 , … , 𝑢𝑛 , 𝑢𝑛+1 , … , 𝑢𝑚 ) = (𝑢1 , … , 𝑢𝑛 ). If we write the restriction of
𝑌 to 𝑉 as 𝑌 = ∑𝑛𝑗=1 𝑌 𝑗 (𝑣)𝜕/𝜕𝑣𝑗 , an easy computation shows that the vector field
∑𝑛𝑗=1 𝑌 𝑗 (𝑢1 , … , 𝑢𝑛 )𝜕/𝜕𝑢𝑗 on 𝑈 is 𝜋-related to 𝑌 . Blending together all of these
vector fields on 𝑀 with a partition of unity yields a global vector field that is 𝜋-
related to 𝑌 .
62 2. Complex Submanifolds

Now given 𝑝 ∈ 𝑁, let (𝑥1 , … , 𝑥𝑛 ) be smooth coordinates centered at 𝑝 on

an open set 𝑈 ⊆ 𝑁 whose image is the unit ball 𝐵1 (0) ⊆ ℝ𝑛 . Given 𝑞 ∈ 𝑈 with
coordinate representation (𝑞 1 , … , 𝑞 𝑛 ), let 𝑊 be the constant-coefficient vector field
𝑞 𝑗 𝜕/𝜕𝑥𝑗 on 𝑈 , and let 𝑌 be the vector field on 𝑁 that is zero outside 𝑈 and equal
to 𝜓𝑊 on 𝑈 , where 𝜓 ∶ 𝑁 → [0, 1] is a smooth bump function that is compactly
supported in 𝑈 and equal to 1 on the line segment from 𝑝 to 𝑞. Because 𝑌 is
compactly supported, it is complete. Let 𝜃 ∶ ℝ × 𝑁 → 𝑁 be its flow. The curve
𝛾(𝑡) = (𝑡𝑞 1 , … , 𝑡𝑞 𝑛 ) satisfies 𝛾 ′ (𝑡) = 𝑌𝛾(𝑡) for 𝑡 ∈ [0, 1], so it is a portion of the
integral curve of 𝑌 starting at 𝑝. Thus the time-1 flow 𝜃1 ∶ 𝑁 → 𝑁 takes 𝑝 to 𝑞.
Let 𝑋 be a lift of 𝑌 to 𝑀. Because 𝜋 is proper, 𝑋 is supported in the compact set
𝜋 −1 (supp 𝑌 ), so it is also complete. Let Θ ∶ ℝ × 𝑀 → 𝑀 be its flow. The fact that
𝑌 and 𝑋 are 𝜋-related implies 𝜃1 ∘𝜋 = 𝜋 ∘Θ1 by [LeeSM, Prop. 9.13]. In particular,
this means that for all 𝑝 ̃ ∈ 𝜋 −1 (𝑝), we have 𝜋(Θ1 (𝑝))
̃ = 𝜃1 (𝜋(𝑝))
̃ = 𝜃1 (𝑝) = 𝑞, so
Θ1 maps 𝜋 (𝑝) to 𝜋 (𝑞). Its inverse Θ−1 maps 𝜋 (𝑞) to 𝜋 (𝑝), showing that the
−1 −1 −1 −1

fibers 𝜋 −1 (𝑝) and 𝜋 −1 (𝑞) are diffeomorphic.

Now define an equivalence relation on 𝑁 by 𝑝1 ∼ 𝑝2 if and only if 𝜋 −1 (𝑝1 ) is
diffeomorphic to 𝜋 −1 (𝑝2 ). The argument above shows that each equivalence class
is open in 𝑁, and since 𝑁 is connected, there is only one equivalence class. □

And here is the commutative algebra result. Proofs can be found in [vdW50,
Section 82] or [GKZ08, Chapter 13].
Lemma 2.42. Suppose 𝑓0 , … , 𝑓𝑛 are homogeneous polynomials in 𝑛 + 1 complex
variables. There is a quantity 𝑅(𝑓0 , … , 𝑓𝑛 ) ∈ ℂ, called the resultant of 𝒇0 , …, 𝒇𝒏 ,
which is given by a homogeneous polynomial in the coefficients of 𝑓0 , … , 𝑓𝑛 , and
which is zero if and only if 𝑓0 , … , 𝑓𝑛 have a common zero other than the origin.
Theorem 2.43. Let 𝑛 and 𝑚 be positive integers, with 𝑛 ≥ 2. All nonsingular
projective hypersurfaces of degree 𝑚 in ℂℙ𝑛 are diffeomorphic to each other.

Proof. Let 𝑛 and 𝑚 be fixed. By Proposition 2.31, each nonsingular projective

hypersurface 𝑆 ⊆ ℂℙ𝑛 of degree 𝑚 is the variety determined by a homogeneous
polynomial 𝑝 of degree 𝑚, and nonsingularity means that 𝑑𝑝𝑤 does not vanish at
any nonzero point such that 𝑝(𝑤) = 0. Any two such polynomials that are nonzero
constant multiples of each other determine the same hypersurface. Conversely, if 𝑝1
and 𝑝2 determine the same nonsingular hypersurface, then the ratio 𝑝1 /𝑝2 extends
holomorphically across the zero set 𝑝−12 (0) to define a global holomorphic function
on ℂ 𝑛+1
∖ {0}, and thus to all of ℂ 𝑛+1
by Hartogs’s extension theorem. Because it
is homogeneous of degree zero, it is constant.

degree 𝑚 by a certain subset of the projective space ℙ(𝐻), where 𝐻 is the complex
Therefore we can parametrize the set of nonsingular projective hypersurfaces of

vector space of homogeneous polynomials of degree 𝑚 in 𝑛 + 1 complex variables.

The variety determined by a nonzero element 𝑝 ∈ 𝐻 is nonsingular if and only if
Complex Submanifolds of Projective Spaces 63

𝑝, 𝜕0 𝑝, … , 𝜕𝑛 𝑝 have no common zeros other than the origin (where 𝜕𝑗 𝑝 is shorthand

for 𝜕𝑝/𝜕𝑤𝑗 ). Because 𝑝 is homogeneous, it satisfies Euler’s identity:
𝜕𝑝
𝑤𝑗 (𝑤) = 𝑚𝑝(𝑤),
𝜕𝑤𝑗
which is proved by differentiating the equation 𝑝(𝜆𝑤) = 𝜆𝑚 𝑝(𝑤) with respect to 𝜆
and setting 𝜆 = 1. Thus if 𝑎 ∈ ℂ𝑛+1 ∖ {0} is a common zero of 𝜕0 𝑝, … , 𝜕𝑛 𝑝, it is
also a zero of 𝑝.
This shows that the nonsingular hypersurfaces of degree 𝑚 are exactly the va-
rieties 𝑉𝑝 for which the homogeneous polynomials 𝜕0 𝑝, … , 𝜕𝑛 𝑝 have no common
zeros other than the origin. Let 𝑊 ⊆ ℙ(𝐻) be the corresponding set. It follows
from Lemma 2.42 that 𝑊 is the complement of the algebraic variety 𝑉𝑅 defined
by a homogeneous polynomial 𝑅. If 𝑉𝑅 were nonsingular, it would follow from
Proposition 2.17 that 𝑊 is path-connected; we can show that the same result holds
even if 𝑉𝑅 has singularities, as follows. Given distinct points [𝑝], [𝑞] ∈ 𝑊 , let
𝐹 ∶ ℂℙ1 → ℙ(𝐻) be the holomorphic embedding 𝐹 ([𝑤0 , 𝑤1 ]) = [𝑤0 𝑝 + 𝑤1 𝑞],
and let 𝐿 denote the projective line 𝐹 (ℂℙ1 ) ⊆ ℙ(𝐻). Then
𝐹 −1 (𝐿 ∩ 𝑉𝑅 ) = {[𝑤0 , 𝑤1 ] ∶ 𝑅(𝑤0 𝑝 + 𝑤1 𝑞) = 0},
which is an algebraic subvariety of ℂℙ1 . It is not all of ℂℙ1 because it does not
contain [0, 1] or [1, 0], so by Proposition 2.30 it is finite. Since ℂℙ1 is homeomor-
phic to 𝕊2 , the complement of a finite set in ℂℙ1 is path-connected, so there is a
path in 𝐿 ∩ 𝑊 from 𝑝 to 𝑞.
Now define a subset 𝑋 ⊆ ℙ(𝐻) × ℂℙ𝑛 by
𝑋 = {([𝑝], [𝑤]) ∶ [𝑝] ∈ ℙ(𝐻) and [𝑤] ∈ 𝑉𝑝 },
and let 𝑋0 = 𝑋 ∩ 𝜋1−1 (𝑊 ), where 𝜋1 ∶ ℙ(𝐻) × ℂℙ𝑛 → ℙ(𝐻) is the projection. Let
Π ∶ 𝑋0 → 𝑊 be the restriction of 𝜋1 . For each [𝑝] ∈ 𝑊 , the preimage Π−1 {[𝑝]}
is the set {[𝑝]} × 𝑉𝑝 , which is biholomorphic to 𝑉𝑝 ; thus the fibers of Π are exactly
the nonsingular degree-𝑚 hypersurfaces in ℂℙ𝑛 .
Note that 𝑋 is the image of the set
̃ = {(𝑝, 𝑤) ∶ 𝑝(𝑤) = 0} ⊆ (𝐻 ∖ {0}) × (ℂ𝑛+1 ∖ {0})
𝑋
under the projection 𝑞 ∶ (𝐻 ∖{0})×(ℂ𝑛+1 ∖{0}) → ℙ(𝐻)×ℂℙ𝑛 given by 𝑞(𝑝, 𝑤) =
([𝑝], [𝑤]). Since 𝑋 ̃ is the zero set of the continuous function 𝑓 (𝑝, 𝑤) = 𝑝(𝑤),
it is closed. Moreover, every point ([𝑝], [𝑤]) ∈ 𝑋 has a representative (𝑝′ , 𝑤′ ) =
(𝑝/|𝑝|, 𝑤/|𝑤|) in the compact set 𝐾 × 𝕊2𝑛+1 ⊆ 𝐻 × ℂ𝑛+1 , where 𝐾 is the unit sphere
in 𝐻 with respect to any choice of norm. Thus 𝑋 is the image of the compact set
𝑋̃ ∩ (𝐾 × 𝕊2𝑛+1 ), so it is compact, and therefore the restriction of 𝜋1 to 𝑋 is proper.
It follows that Π ∶ 𝑋0 → 𝑊 is also proper because it is the restriction of 𝜋1 |𝑋 to an
open subset 𝑋0 ⊆ 𝑋 that is saturated, meaning it is the full preimage of a subset
of ℙ(𝐻) (see [LeeTM, Prop. 4.93(f)]).
64 2. Complex Submanifolds

Now we show that 𝑋0 is a codimension-1 complex submanifold of ℙ(𝐻)×ℂℙ𝑛 .

Given any point 𝑥0 = ([𝑝0 ], [𝑤0 ]) ∈ 𝑋0 , we can choose a neighborhood 𝑈 on
which we have affine coordinates (𝑞 1 , … , 𝑞 𝑁 ) for ℙ(𝐻) (where 𝑁 = dim ℙ(𝐻))
and affine coordinates (𝑧1 , … , 𝑧𝑛 ) for ℂℙ𝑛 , and then the set 𝑋0 ∩ 𝑈 is characterized
by a polynomial equation 𝑓 (𝑞, 𝑧) = 0 which is the coordinate representation of
𝑝(𝑤) = 0. The fact that [𝑝0 ] ∈ 𝑊 guarantees that some 𝑧𝑗 -derivative of 𝑓 does
not vanish at 𝑥0 , so 𝑋0 ∩ 𝑈 is a regular level set of a single holomorphic function,
showing that 𝑋0 is a complex submanifold of dimension 𝑁 + 𝑛 − 1. Moreover, the
fact that some 𝑧𝑗 -derivative of 𝑓 is nonzero means that the tangent space 𝑇𝑥0 𝑋0
does not contain all of the direct summand 𝑇[𝑧0 ] ℂℙ𝑛 ⊆ 𝑇𝑥0 (ℙ(𝐻) × ℂℙ𝑛 ); thus the
dimension of Ker 𝑑Π𝑥0 = 𝑇[𝑧0 ] ℂℙ𝑛 ∩ 𝑇𝑥0 𝑋0 is less than 𝑛. It follows that the rank
of 𝑑Π𝑥0 is at least (𝑁 + 𝑛 − 1) − (𝑛 − 1) = 𝑁, so Π is a submersion. Because 𝑊 is
connected, it follows from Lemma 2.41 that all of the fibers of Π are diffeomorphic
to each other. □

The Holomorphic Embedding Problem

One of the fundamental results in smooth manifold theory is the Whitney embed-
ding theorem, which says that every smooth manifold can be properly embedded in
some Euclidean space; see [LeeSM, Thm. 6.15] for a proof. (Properness of the em-
bedding is equivalent to the image being a closed subset; see [LeeSM, Prop. 5.5].)
In complex manifold theory, things are very different. In the first place, we

positive dimension into ℂ𝑛 (see Example 2.25). It turns out that the appropriate
cannot hope to find holomorphic embeddings of compact complex manifolds of

place to look for embeddings of compact complex manifolds is in ℂℙ𝑛 , but not

characterizing which compact manifolds admit embeddings into ℂℙ𝑛 will occupy
every such manifold admits such an embedding even in that case. The question of

much of the rest of this book.

What about noncompact complex manifolds? Here again, the situation is not

plex manifold 𝑀 to be embeddable in ℂ𝑛 . For example, the preceding observation

straightforward. There are nontrivial necessary conditions for a noncompact com-

implies that if 𝑀 contains a positive-dimensional compact complex submanifold,

then 𝑀 cannot be embedded in ℂ𝑛 . It turns out that the class of complex manifolds
that can be properly embedded in ℂ𝑛 for some 𝑛 has a nice intrinsic characterization.

Let 𝑀 be a complex manifold and 𝒪(𝑀) its ring of global holomorphic func-
First, a few definitions.

tions. We say 𝒪(𝑀) separates points if for every pair of distinct points 𝑝, 𝑞 ∈ 𝑀,
there exists 𝑓 ∈ 𝒪(𝑀) that satisfies 𝑓 (𝑝) = 0 and 𝑓 (𝑞) ≠ 0. We say 𝒪(𝑀) sep-
arates directions if for every 𝑝 ∈ 𝑀 and every nonzero 𝑣 ∈ 𝑇𝑝′ 𝑀, there exists
𝑓 ∈ 𝒪(𝑀) such that 𝑣𝑓 ≠ 0. The significance of the latter condition is explained
by the following lemma.
The Holomorphic Embedding Problem 65

Lemma 2.44. For a complex 𝑛-manifold 𝑀, 𝒪(𝑀) separates directions if and only
if for each 𝑝 ∈ 𝑀 there exist global holomorphic functions 𝑧1 , … , 𝑧𝑛 ∈ 𝒪(𝑀) that
restrict to local holomorphic coordinates in a neighborhood of 𝑝.

Proof. On the one hand, if there exist such functions 𝑧1 , … , 𝑧𝑛 , then for any
nonzero vector 𝑣 = 𝑣𝑗 𝜕/𝜕𝑧𝑗 |𝑝 ∈ 𝑇𝑝′ 𝑀, one of the components 𝑣𝑗 must be nonzero,
and 𝑣(𝑧𝑗 ) = 𝑣𝑗 ≠ 0.
Conversely, suppose 𝒪(𝑀) separates directions, and let 𝑝 ∈ 𝑀 be arbi-
trary. We will show by induction on 𝑘 that for each 𝑘 = 1, … , 𝑛, there exist
𝑧1 , … , 𝑧𝑘 ∈ 𝒪(𝑀) such that 𝑑𝑧1 |𝑝 , … , 𝑑𝑧𝑘 |𝑝 are linearly independent. If this is
true for 𝑘 = 𝑛, the holomorphic inverse function theorem shows that (𝑧1 , … , 𝑧𝑛 )
restrict to holomorphic coordinates in a neighborhood of 𝑝.
For 𝑘 = 1, just choose any 𝑧1 ∈ 𝒪(𝑀) such that 𝑣(𝑧1 ) ≠ 0 for some nonzero
𝑣 ∈ 𝑇𝑝′ 𝑀, which implies 𝑑𝑧1 |𝑝 ≠ 0. Now suppose the claim is true for some 𝑘 < 𝑛.
The complex-linear map from 𝑇𝑝′ 𝑀 to ℂ𝑘 given by 𝑣 ↦ (𝑑𝑧1 |𝑝 (𝑣), … , 𝑑𝑧𝑘 |𝑝 (𝑣))
has rank 𝑘, so it has a kernel of dimension 𝑛 − 𝑘. Choose 𝑣 ≠ 0 in that kernel, and
let 𝑧𝑘+1 ∈ 𝒪(𝑀) be a function such that 𝑣(𝑧𝑘+1 ) ≠ 0. It follows that 𝑑𝑧𝑘+1 |𝑝 is
linearly independent of 𝑑𝑧1 |𝑝 , … , 𝑑𝑧𝑘 |𝑝 , thus completing the induction. □

One last definition: for any subset 𝐾 ⊆ 𝑀, define the holomorphic hull of 𝑲
as the set
̂ = 𝑧 ∈ 𝑀 ∶ |𝑓 (𝑧)| ≤ sup |𝑓 | for all 𝑓 ∈ 𝒪(𝑀) .
𝐾 { 𝐾
}

(The terminology is motivated by analogy with the convex hull of a subset 𝐾 ⊆ ℝ𝑛 ,

which is the intersection of all convex subsets of ℝ𝑛 containing 𝐾. Problem 2-12

functions instead of holomorphic ones.) We say 𝑀 is holomorphically convex if

shows that the convex hull can be characterized in an analogous way using linear

whenever 𝐾 is a compact subset of 𝑀, 𝐾 ̂ is also compact.

A complex manifold 𝑀 is called a Stein manifold if it satisfies the following
three conditions:
(i) 𝒪(𝑀) separates points.
(ii) 𝒪(𝑀) separates directions.
(iii) 𝑀 is holomorphically convex.
Proposition 2.45. Suppose 𝑀 ⊆ ℂ𝑛 is a properly embedded complex submanifold.
Then 𝑀 is Stein.

Proof. Clearly 𝒪(𝑀) separates points: given distinct points 𝑝, 𝑞 ∈ 𝑀, some holo-
morphic coordinate function 𝑧𝑗 takes on different values at those two points, and
then the restriction to 𝑀 of 𝑓 (𝑧) = 𝑧𝑗 − 𝑝𝑗 is zero at 𝑝 and nonzero at 𝑞.
66 2. Complex Submanifolds

Similarly, 𝒪(𝑀) separates directions: given 𝑝 ∈ 𝑀 and 𝑣 ≠ 0 ∈ 𝑇𝑝′ 𝑀, we can

consider 𝑣 as an element of 𝑇𝑝′ ℂ𝑛 and write 𝑣 = 𝑣𝑗 𝜕/𝜕𝑧𝑗 |𝑝 . Some component 𝑣𝑗 is
nonzero, from which it follows that 𝑣(𝑧𝑗 ) = 𝑣𝑗 ≠ 0.
To show that 𝑀 is holomorphically convex, let 𝐾 ⊆ 𝑀 be a compact subset and
̂
let 𝐾 ⊆ 𝑀 be its holomorphic hull. First note that 𝐾 ̂ is closed in 𝑀: if 𝑝 ∈ 𝑀 is a
boundary point of 𝐾,̂ there is a sequence of points 𝑝𝑗 in 𝐾 such that 𝑝𝑗 → 𝑝. Then
for any 𝑓 ∈ 𝒪(𝑀), since |𝑓 (𝑝𝑗 )| ≤ sup𝐾 |𝑓 | for each 𝑗, it follows by continuity
̂ Since 𝑀 is closed in ℂ𝑛 , it follows that 𝐾
that |𝑓 (𝑝)| ≤ sup𝐾 |𝑓 |, so 𝑝 ∈ 𝐾. ̂ is also
closed in ℂ .𝑛

Next we show 𝐾 ̂ is bounded. Because 𝐾 is compact, there is some 𝑟 > 0 such

that 𝐾 is contained in the closed ball 𝐵 𝑟 (0) ⊆ ℂ𝑛 . Suppose 𝑤 ∈ 𝑀 ∖ 𝐵 𝑟 (0). Set
𝑅 = |𝑤| > 𝑟, and let 𝑓 ∶ 𝑀 → ℂ be the restriction of the holomorphic linear
function 𝑓 (𝑧) = 𝑧 ⋅ 𝑤 = 𝑧1 𝑤1 + ⋯ + 𝑧𝑛 𝑤𝑛 . Now 𝑓 (𝑤) = |𝑤|2 = 𝑅2 ; but if 𝑧 ∈ 𝐾,

|𝑓 (𝑧)| = |𝑧 ⋅ 𝑤| ≤ |𝑧| |𝑤| ≤ 𝑟𝑅 < 𝑅2 = |𝑓 (𝑤)|.

then the Cauchy–Schwartz inequality gives

̂ The contrapositive of this statement is

This shows |𝑓 (𝑤)| > sup𝐾 |𝑓 |, so 𝑤 ∉ 𝐾.
̂ ⊆ 𝐵 𝑟 (0). Since 𝐾
𝐾 ̂ is closed and bounded in ℂ𝑛 , it is compact. □

The most important fact about Stein manifolds is the following converse to the
preceding proposition, proved in 1961 by Errett Bishop and Raghavan Narasimhan.
We will neither prove nor use it, but you can find a proof in [Hör90] or [GR09].

proper holomorphic embedding into ℂ𝑛 for some 𝑛.

Theorem 2.46 (Stein Embedding Theorem). Every Stein manifold admits a

It is easy to come up with examples of noncompact complex manifolds that are

not Stein.
Example 2.47 (Non-Stein Manifolds).
(a) Suppose 𝑀 is a complex manifold that contains a compact complex sub-
manifold 𝑆 ⊆ 𝑀 of positive dimension. Then every 𝑓 ∈ 𝒪(𝑀) restricts
to a constant function on each connected component of 𝑆, so 𝒪(𝑀) does
not separate points. Thus, for example, no compact complex manifold of
positive dimension is Stein, and no product of such a manifold with any

(b) Let 𝑈 be an arbitrary open subset of ℂ𝑛 with 𝑛 ≥ 2, 𝑝 ∈ 𝑈 , and 𝑀 = 𝑈 ∖

other complex manifold is Stein.

{𝑝}. We will show that 𝑀 is not holomorphically convex, and therefore

not Stein. Choose 𝜀 > 0 such that 𝐵 𝜀 (𝑝) ⊆ 𝑈 , and let 𝐾 = 𝜕𝐵𝜀 (𝑝).
Suppose 𝑓 ∈ 𝒪(𝑀) is arbitrary. Then the Hartogs extension theorem
shows that 𝑓 has a holomorphic extension (still denoted by 𝑓 ) to all of
𝑈 . The restriction of |𝑓 | to the compact set 𝐵 𝜀 (𝑝) achieves a maximum,
and the maximum principle guarantees that this maximum occurs on 𝐾.
The Holomorphic Embedding Problem 67

It follows that for every 𝑧 ∈ 𝐵 𝜀 (𝑝) ∩ 𝑀, we have |𝑓 (𝑧)| ≤ sup𝐾 |𝑓 |, so

𝐵 𝜀 (𝑝) ∩ 𝑀 ⊆ 𝐾.̂ But any sequence in 𝐵 𝜀 (𝑝) ∩ 𝑀 that converges to 𝑝 has
no convergent subsequence in 𝐾, ̂ so 𝐾 ̂ is not compact. //
On the other hand, some open submanifolds of ℂ𝑛 are Stein (in addition to
ℂ𝑛 itself). If 𝑈 ⊆ ℂ𝑛 is open, using the restrictions of the coordinate functions
𝑧1 , … , 𝑧𝑛 to 𝑈 , we can show that 𝒪(𝑈 ) separates points and directions as in the
proof of Proposition 2.45, so an open submanifold of ℂ𝑛 is Stein if and only if it is
holomorphically convex. Here is a class of examples for which we can show that is
the case.
Example 2.48 (Convex Open Submanifolds Are Stein). Suppose 𝑈 ⊆ ℂ𝑛 is a
convex open subset. To see that 𝑈 is holomorphically convex, let 𝐾 ⊆ 𝑈 be an
arbitrary compact subset. We begin by showing that the holomorphic hull 𝐾 ̂ is
contained in the convex hull of 𝐾 (considered as a subset of ℝ ), which we denote
2𝑛

by ch(𝐾). Suppose 𝑝 ∈ 𝑈 ∖ ch(𝐾). It follows from Problem 2-12 that there exists
a real-linear function 𝑓 ∶ ℂ𝑛 → ℝ such that 𝑓 (𝑝) > sup𝐾 𝑓 . Define 𝛼 ∶ ℂ𝑛 → ℂ
by 𝛼(𝑧) = 𝑓 (𝑧) − 𝑖𝑓 (𝑖𝑧); a simple computation shows that 𝛼(𝑖𝑧) = 𝑖𝛼(𝑧), so 𝛼
is a complex-linear functional whose real part is 𝑓 . Then 𝑧 ↦ 𝑒𝛼(𝑧) restricts to a
holomorphic function on 𝑈 , which satisfies

|𝑒 |=𝑒 > sup 𝑒𝑓 = sup |𝑒𝛼 | ,

𝛼(𝑝) 𝑓 (𝑝)
𝐾 𝐾
̂
so 𝑝 ∉ 𝐾.
Since 𝐾 is compact, it is easy to verify that ch(𝐾) is also compact, and since
𝑈 is convex, ch(𝐾) is contained in 𝑈 . Now 𝐾 ̂ is closed in 𝑈 by continuity, so it is
a closed subset of the compact set ch(𝐾) and thus compact. //

submanifolds of ℂ𝑛 . But convexity is not a necessary condition, for the simple

Convexity is an easy way to recognize some Stein manifolds among the open

manifold is. For example, consider ℂ2 with holomorphic coordinates (𝑧, 𝑤). The
reason that it is not biholomorphically invariant, while the condition of being a Stein

open subset 𝑈 = {(𝑧, 𝑤) ∶ Re 𝑤 > 0} is convex and thus is a Stein manifold. But
now consider the holomorphic map 𝜑 ∶ ℂ2 → ℂ2 given by 𝜑(𝑧, 𝑤) = (𝑧, 𝑤 + 𝑧2 ).
It is a biholomorphism with inverse 𝜑−1 (𝑧, 𝑤) = (𝑧, 𝑤 − 𝑧2 ). Thus 𝜑(𝑈 ) is also
a Stein manifold. But 𝜑(𝑈 ) is the subset consisting of all (𝑧, 𝑤) such that Re 𝑤 >
(Re 𝑧)2 − (Im 𝑧)2 , which is not convex.
There is a biholomorphically invariant property called pseudoconvexity (which

subset of ℂ𝑛 is a Stein manifold if and only if it is pseudoconvex. Since the deep

is implied by convexity but more general), and it can be shown that a connected open

properties of Stein manifolds and pseudoconvex domains belong more properly to

the theory of complex analysis in several variables, we will have no more to say
about them in this book; but you can find a discussion of these matters in [Kra01],
starting in Chapter 3.
68 2. Complex Submanifolds

Problems
2-1. For each 𝑛 ≥ 1, let U(𝑛) ⊆ GL(𝑛, ℂ) be the 𝑛-dimensional unitary group,
that is, the subgroup of matrices 𝐴 ∈ GL(𝑛, ℂ) that satisfy 𝐴∗ 𝐴 = Id.
(Here 𝐴∗ denotes the Hermitian adjoint of 𝑨, that is, the transposed con-
jugate of 𝐴.) Show that U(𝑛) is not a complex submanifold of GL(𝑛, ℂ).
2-2. Prove Lemma 2.27 (a subset of ℂℙ𝑛 is a complex submanifold if and only
its preimage in ℂ𝑛+1 ∖ {0} is a complex submanifold.)
2-3. For positive integers 𝑚, 𝑛, let 𝑆 ∶ ℂℙ𝑚 × ℂℙ𝑛 → ℂℙ𝑚𝑛+𝑚+𝑛 be the map

𝑆([𝑤0 , … , 𝑤𝑚 ], [𝑧0 , … , 𝑧𝑛 ])
= [𝑤0 𝑧0 , 𝑤0 𝑧1 , … , 𝑤0 𝑧𝑛 , 𝑤1 𝑧0 , … , 𝑤1 𝑧𝑛 , … , 𝑤𝑚 𝑧0 , … , 𝑤𝑚 𝑧𝑛 ],

Show that 𝑆 is a holomorphic embedding, called the Segre embedding.

where the products on the right-hand side are listed in lexicographic order.

Use this to conclude that every product of projective complex manifolds

is projective.

phism 𝐹 ∶ 𝕊2 → ℂℙ1 :
2-4. Prove that the following formulas determine a well-defined diffeomor-

[𝑥 + 𝑖𝑦, 1 − 𝑧], (𝑥, 𝑦, 𝑧) ≠ (0, 0, 1),

𝐹 (𝑥, 𝑦, 𝑧) =
{[1 + 𝑧, 𝑥 − 𝑖𝑦], (𝑥, 𝑦, 𝑧) ≠ (0, 0, −1).

2-5. Show that every nonsingular quadric in ℂℙ2 is biholomorphic to ℂℙ1 .

tion of the affine curve 𝑤 = 𝑧2 .]

[Hint: Start by thinking about how to parametrize the projective comple-

2-6. Show that every nonsingular quadric in ℂℙ3 is biholomorphic to ℂℙ1 ×

ℂℙ1 . [Hint: Use Problem 2-3.]
2-7. For 𝑛 ≥ 3, let G+ 2
(ℝ𝑛 ) be the Grassmannian of oriented 2-planes in ℝ𝒏 ,
that is, the set of 2-dimensional linear subspaces of ℝ𝑛 together with a
choice of orientation for each. The natural action of SO(𝑛) on ℝ𝑛 induces
a transitive action on G+ 2
(ℝ𝑛 ) with a closed isotropy group, giving it a
smooth manifold structure as a homogeneous space by [LeeSM, Thm.

lar quadric in ℂℙ𝑛−1 is diffeomorphic to G+ (ℝ𝑛 ). [Hint: Consider the

21.20]. (You do not need to prove this.) Prove that every nonsingu-
2
quadric 𝑄 defined by (𝑤1 )2 + ⋯ + (𝑤𝑛 )2 = 0, and show that the map
𝑄 → G+ 2
(ℝ𝑛 ) by [𝑤] ↦ span(Re 𝑤, Im 𝑤) is an SO(𝑛)-equivariant diffeo-
morphism.]
2-8. Prove that every Fermat hypersurface (Example 2.39) is nonsingular.
Problems 69

(a) Show that every automorphism of ℂℙ1 is a projective transformation.

[Hint: First show that it suffices to assume 𝐹 ([1, 0]) = [1, 0] and
2-9.

𝐹 ([0, 1] = [0, 1].]

(b) Show that every automorphism of ℂ is an affine function of the form
𝑓 (𝑧) = 𝑎𝑧 + 𝑏 for some 𝑎, 𝑏 ∈ ℂ. [Hint: Begin by extending 𝑓
to a bijection 𝐹 ∶ ℂℙ1 → ℂℙ1 , and show that 𝐹 has a removable

2-10. Show that every holomorphic map from ℂℙ1 to itself can be written in
singularity at infinity.]

the form 𝐹 ([𝑧, 𝑤]) = [𝑝(𝑧, 𝑤), 𝑞(𝑧, 𝑤)], where 𝑝 and 𝑞 are homogeneous

2-11. Suppose 𝑀 is a disconnected compact complex manifold. Show that 𝑀

polynomials of the same degree with no common zeros except the origin.

2-12. Suppose 𝐾 is an arbitrary subset of ℝ𝑛 . The convex hull of 𝑲 is the inter-

is projective if and only if each of its connected components is projective.

section of all convex subsets of ℝ𝑛 containing 𝐾. Show that the convex

hull of 𝐾 is equal to the set

{𝑥 ∈ ℝ ∶ 𝑓 (𝑥) ≤ sup 𝑓 for all linear functions 𝑓 ∶ ℝ𝑛 → ℝ}.

𝑛
𝐾
2-13. Let 𝑉 be a finite-dimensional complex vector space. Show that each
Grassmannian G𝑘 (𝑉 ) is projective. [Hint: Let Λ𝑘 𝑉 be the set of alternat-
ing contravariant 𝑘-tensors on 𝑉 , and define a map from G𝑘 (𝑉 ) to ℙ(Λ𝑘 𝑉 )
by sending the span of (𝑣1 , … , 𝑣𝑘 ) to [𝑣1 ∧ ⋯ ∧ 𝑣𝑘 ]. Show that this is a

2-14. Let ℍ = {𝑤 ∈ ℂ ∶ Im 𝑤 > 0} be the upper half-plane. Define an action

holomorphic embedding, called a Plücker embedding.]

of ℤ2 on ℂ × ℍ by (𝑚, 𝑛) ⋅ (𝑧, 𝑤) = (𝑧 + 𝑚 + 𝑛𝑤, 𝑤).

(a) Show that this action is free, proper, and holomorphic, so 𝑀 = (ℂ ×
ℍ)/ℤ2 is a complex 2-manifold.
(b) Show that the projection 𝜋2 ∶ ℂ × ℍ → ℍ descends to a surjective
holomorphic submersion Π ∶ 𝑀 → ℍ.
(c) Show that every 1-dimensional complex torus is biholomorphic to at
least one of the fibers of Π. [Hint: Use Problem 1-4.]
 10.1090/gsm/244/03

Chapter 3

Holomorphic Vector
Bundles

bundle 𝜋 ∶ 𝐸 → 𝑀 in which 𝐸 and 𝑀 are complex manifolds, 𝜋 is a holomorphic

Recall that in Chapter 1 we defined a holomorphic vector bundle as a complex vector

map, and the local trivializations can be chosen to be biholomorphisms. In this

chapter we will delve much more deeply into the theory of holomorphic vector
bundles.

Holomorphic Bundle Tools

Most of the basic tools for working with smooth vector bundles have direct ana-
logues in the holomorphic category. Here we summarize the main ones.

Lemma 3.1 (Transition Functions of Holomorphic Bundles). Suppose 𝜋 ∶ 𝐸 →

𝑀 is a holomorphic vector bundle of rank 𝑘, and Φ𝛼 ∶ 𝜋 −1 (𝑈𝛼 ) → 𝑈𝛼 × ℂ𝑘 and
Φ𝛽 ∶ 𝜋 −1 (𝑈𝛽 ) → 𝑈𝛽 ×ℂ𝑘 are two holomorphic local trivializations with 𝑈𝛼 ∩𝑈𝛽 ≠
∅. Then for all (𝑝, 𝑣) ∈ (𝑈𝛼 ∩ 𝑈𝛽 ) × ℂ𝑘 , we have

(3.1) Φ𝛼 ∘ Φ−1
𝛽 (𝑝, 𝑣) = (𝑝, 𝜏𝛼𝛽 (𝑝)𝑣)

for some holomorphic map 𝜏𝛼𝛽 ∶ 𝑈𝛼 ∩ 𝑈𝛽 → GL(𝑘, ℂ) called the transition func-
tion from 𝚽𝜷 to 𝚽𝜶 .

Proof. Exactly the same as the corresponding proof for smooth bundles, with
“holomorphic” substituted for “smooth”; see [LeeSM, Lemma 10.5]. □

71
72 3. Holomorphic Vector Bundles

For a holomorphic vector bundle 𝐸 → 𝑀, a local or global section is called a

terms of any holomorphic local frame (𝑠1 , … , 𝑠𝑘 ) over an open set 𝑈 ⊆ 𝑀, we can
holomorphic section if it is holomorphic as a map between complex manifolds. In

write a rough section 𝜎 locally as 𝜎(𝑝) = 𝜎 𝑗 (𝑝)𝑠𝑗 (𝑝) = 𝜎 1 (𝑝)𝑠1 (𝑝) + ⋯ + 𝜎 𝑘 (𝑝)𝑠𝑘 (𝑝)
for some complex-valued functions 𝜎 1 , … , 𝜎 𝑘 ∶ 𝑈 → ℂ, called the component
functions of 𝝈 with respect to the given frame. We let 𝒪(𝑈 ; 𝐸) denote the complex
vector space of local holomorphic sections of 𝐸 over 𝑈 , so that 𝒪(𝑀; 𝐸) is the
space of global holomorphic sections.
Lemma 3.2 (Local Frame Criterion for Holomorphicity). Let 𝐸 → 𝑀 be a
holomorphic vector bundle. Given a rough local section 𝜎 ∶ 𝑈 → 𝐸 and a holo-
morphic local frame for 𝐸 over 𝑈 , the section 𝜎 is holomorphic on 𝑈 if and only if
its component functions are holomorphic.

Proof. Just the same as the corresponding proof for smooth bundles [LeeSM,
Prop. 10.22]. □

morphic vector bundle 𝜋 ∶ 𝐸 → 𝑀, there is a one-to-one correspondence be-

Lemma 3.3 (Holomorphic Local Frames and Local Trivializations). For a holo-

cal frame (𝑠1 , … , 𝑠𝑘 ) corresponding to a local trivialization (𝑈 , Φ) is defined by

tween holomorphic local frames and holomorphic local trivializations. The lo-

𝑠𝑗 (𝑝) = Φ−1 (𝑝, 𝑒𝑗 ), where 𝑒𝑗 is the 𝑗th standard basis vector for ℂ𝑘 .

Proof. Again, just like its smooth counterpart [LeeSM, Prop. 10.19]. □
Lemma 3.4 (Holomorphic Vector Bundle Chart Lemma). Suppose 𝑀 is a com-
plex manifold, and for each 𝑝 ∈ 𝑀 we are given a 𝑘-dimensional complex vec-
tor space 𝐸𝑝 . Let 𝐸 = ∐𝑝∈𝑀 𝐸𝑝 (the disjoint union of the spaces 𝐸𝑝 ), and let
𝜋 ∶ 𝐸 → 𝑀 be the obvious projection. Suppose further that we are given
(i) an indexed open cover {𝑈𝛼 }𝛼∈𝐴 of 𝑀;
(ii) for each 𝛼 ∈ 𝐴, a bijection Φ𝛼 ∶ 𝜋 −1 (𝑈𝛼 ) → 𝑈𝛼 × ℂ𝑘 whose restriction
to each 𝐸𝑝 is a complex-linear isomorphism from 𝐸𝑝 to {𝑝} × ℂ𝑘 ;
(iii) for each 𝛼, 𝛽 ∈ 𝐴 with 𝑈𝛼 ∩ 𝑈𝛽 ≠ ∅, a holomorphic map 𝜏𝛼𝛽 ∶ 𝑈𝛼 ∩
𝑈𝛽 → GL(𝑘, ℂ) such that Φ𝛼 ∘ Φ−1
𝛽 (𝑝, 𝑣) = (𝑝, 𝜏𝛼𝛽 (𝑝)𝑣) for all (𝑝, 𝑣) ∈
(𝑈𝛼 ∩ 𝑈𝛽 ) × ℂ .
𝑘

Then 𝜋 ∶ 𝐸 → 𝑀 has a unique structure as a holomorphic vector bundle with

{(𝑈𝛼 , Φ𝛼 )} as holomorphic local trivializations, and with 𝜏𝛼𝛽 as the transition func-
tion from Φ𝛽 to Φ𝛼 for each 𝛼 and 𝛽.

Proof. Just like its smooth analogue [LeeSM, Lemma 10.6]. □

spaces 𝐸𝑝 that are given in advance. The next important proposition gives a way
In the vector bundle chart lemma, we construct a vector bundle from vector
Holomorphic Bundle Tools 73

to construct a holomorphic vector bundle out of thin air, given only its transition
functions. To motivate the hypothesis in the following proposition, observe that it

Φ𝛼 , Φ𝛽 , Φ𝛾 overlap, the transition functions satisfy the equation

follows from equation (3.1) that whenever the domains of three local trivializations

𝜏𝛼𝛽 (𝑝)𝜏𝛽𝛾 (𝑝) = 𝜏𝛼𝛾 (𝑝)

for all 𝑝 in the common domain, where the juxtaposition on the left-hand side rep-
resents matrix multiplication. This equation is called the cocycle condition, for
reasons that will become clear in Chapter 6 (see Example 6.3).
Proposition 3.5 (Holomorphic Vector Bundle Construction Theorem). Let 𝑀
be a complex manifold and let 𝒰 = {𝑈𝛼 }𝛼∈𝐴 be an indexed open cover of 𝑀.
Suppose for each 𝛼, 𝛽 ∈ 𝐴 such that 𝑈𝛼 ∩ 𝑈𝛽 ≠ ∅ we are given a holomorphic map
𝜏𝛼𝛽 ∶ 𝑈𝛼 ∩ 𝑈𝛽 → GL(𝑘, ℂ) such that the following cocycle condition is satisfied for
all 𝛼, 𝛽, 𝛾 ∈ 𝐴:

(3.2) 𝜏𝛼𝛽 (𝑝)𝜏𝛽𝛾 (𝑝) = 𝜏𝛼𝛾 (𝑝), 𝑝 ∈ 𝑈𝛼 ∩ 𝑈𝛽 ∩ 𝑈𝛾 .

Then there is a rank-𝑘 holomorphic vector bundle 𝐸 → 𝑀 that admits a holomor-

phic trivialization over each set 𝑈𝛼 , with transition functions given by 𝜏𝛼𝛽 .

Proof. Let 𝐸 ̃ be the disjoint union ∐

𝛼∈𝐴 (𝑈𝛼 × ℂ ), with a point (𝑝, 𝑣) ∈ 𝑈𝛼 × ℂ
𝑘 𝑘

denoted in the disjoint union by (𝑝, 𝑣, 𝛼). Define a relation on 𝐸 ̃ by (𝑝, 𝑣, 𝛼) ∼

(𝑝 , 𝑣 , 𝛼 ) if 𝑝 = 𝑝 and 𝑣 = 𝜏𝛼𝛼 ′ (𝑝)𝑣. By taking two or three of the indices equal
′ ′ ′ ′ ′

to each other, we see that the relation (3.2) implies also that 𝜏𝛽𝛼 (𝑝) = 𝜏𝛼𝛽 (𝑝)−1 and
𝜏𝛼𝛼 (𝑝) = Id, which shows that ∼ is an equivalence relation. Let 𝐸 denote the set
of equivalence classes, and define 𝜋 ∶ 𝐸 → 𝑀 by 𝜋[(𝑝, 𝑣, 𝛼)] = 𝑝; the definition
of the equivalence relation shows that this is well defined. For each 𝑝 ∈ 𝑀, the
fiber 𝐸𝑝 = 𝜋 −1 (𝑝) is the set of all equivalence classes of the form [(𝑝, 𝑣, 𝛼)] for 𝑣
arbitrary and 𝛼 such that 𝑝 ∈ 𝑈𝛼 . We can define a complex vector space structure
on 𝐸𝑝 by choosing a fixed 𝑈𝛼 containing 𝑝 and setting 𝑐1 [(𝑝, 𝑣1 , 𝛼)]+𝑐2 [(𝑝, 𝑣2 , 𝛼)] =
[(𝑝, 𝑐1 𝑣1 + 𝑐2 𝑣2 , 𝛼)] for 𝑐1 , 𝑐2 ∈ ℂ; the fact that the maps 𝑣 ↦ 𝜏𝛼𝛽 (𝑝)𝑣 are all linear
isomorphisms guarantees that this is independent of the choice of 𝛼.
Now for each 𝛼, define a map Φ𝛼 ∶ 𝜋 −1 (𝑈𝛼 ) → 𝑈𝛼 × ℂ𝑘 by Φ𝛼 [(𝑝, 𝑣, 𝛼)] =
(𝑝, 𝑣). It is then straightforward to check that these maps satisfy all of the hypotheses
of the chart lemma and thus define a holomorphic vector bundle structure on 𝐸. □

► Exercise 3.6. Complete the proof of this proposition by verifying that the
chart lemma hypotheses are satisfied.

Just as the preceding proposition allows us to construct a bundle given only the
transition functions, the next one allows us to detect when two bundles are isomor-
phic from the same data.
74 3. Holomorphic Vector Bundles

pose 𝐸 → 𝑀 and 𝐸 ′ → 𝑀 are holomorphic rank-𝑘 vector bundles that both have
Proposition 3.7 (Holomorphic Vector Bundle Isomorphism Criterion). Sup-

holomorphic local trivializations {(𝑈𝛼 , Φ𝛼 )}𝛼∈𝐴 and {(𝑈𝛼 , Φ′𝛼 )}𝛼∈𝐴 over the same
trivializing cover, with transition functions 𝜏𝛼𝛽 and 𝜏𝛼𝛽
′
, respectively. Then 𝐸 and
𝐸 are isomorphic over 𝑀 if and only if for each 𝛼 ∈ 𝐴 there exists a holomorphic
′

map 𝜓𝛼 ∶ 𝑈𝛼 → GL(𝑘, ℂ) such that for 𝛼, 𝛽 ∈ 𝐴 we have

(3.3) 𝜏𝛼𝛽 (𝑝) = 𝜓𝛼 (𝑝)−1 𝜏𝛼𝛽
′
(𝑝)𝜓𝛽 (𝑝), 𝑝 ∈ 𝑈𝛼 ∩ 𝑈𝛽 .

Proof. First suppose there exist such maps 𝜓𝛼 . For each 𝛼, define a holomorphic
bundle homomorphism Ψ𝛼 ∶ 𝑈𝛼 × ℂ𝑘 → 𝑈𝛼 × ℂ𝑘 by
Ψ𝛼 (𝑝, 𝑣) = (𝑝, 𝜓𝛼 (𝑝)𝑣);
it has a holomorphic inverse given by replacing 𝜓𝛼 (𝑝) with 𝜓𝛼 (𝑝)−1 . Then let
𝐹𝛼 ∶ 𝐸|𝑈𝛼 → 𝐸 ′ |𝑈𝛼 be the holomorphic bundle isomorphism 𝐹𝛼 = Φ′𝛼 −1 ∘ Ψ𝛼 ∘ Φ𝛼 .

sition functions satisfy Φ𝛼 ∘ Φ−1

𝛽 (𝑝, 𝑣) = (𝑝, 𝜏𝛼𝛽 (𝑝)𝑣) for 𝑝 ∈ 𝑈𝛼 ∩ 𝑈𝛽 , with a similar
To see what happens when two of these definitions overlap, recall that the tran-

formula for 𝜏𝛼𝛽 . Thus (3.3) implies the following formula on (𝑈𝛼 ∩ 𝑈𝛽 ) × ℂ𝑘 :
′

𝛽 = Ψ 𝛼 ∘ Φ𝛼 ∘ Φ𝛽
Φ𝛼 ∘ Φ−1 −1 ′ ′ −1
∘ Ψ𝛽 .
It follows that 𝐹𝛼 and 𝐹𝛽 agree where both are defined, so they piece together to
create a holomorphic bundle homomorphism 𝐹 ∶ 𝐸 → 𝐸 ′ . It is an isomorphism
because its inverse can be constructed in the same way using 𝜓𝛼 (𝑝)−1 in place of
𝜓𝛼 (𝑝).
Conversely, if 𝐹 ∶ 𝐸 → 𝐸 ′ is a bundle isomorphism, for each 𝛼 ∈ 𝐴 the
composite map Ψ𝛼 = Φ′𝛼 ∘𝐹 ∘Φ−1 𝛼 is a bundle isomorphism from 𝑈𝛼 ×ℂ to itself, so
𝑘

it has the form Ψ𝛼 (𝑝, 𝑣) = (𝑝, 𝜎𝛼 (𝑝, 𝑣)) for some holomorphic map 𝜎𝛼 ∶ 𝑈𝛼 × ℂ𝑘 →
ℂ𝑘 . Since 𝑣 ↦ 𝜎𝛼 (𝑝, 𝑣) is a complex-linear isomorphism for each 𝑝, there is some
function 𝜓𝛼 ∶ 𝑈𝛼 → GL(𝑘, ℂ) such that 𝜎𝛼 (𝑝, 𝑣) = 𝜓𝛼 (𝑝)𝑣. If we let (𝑒𝑗 ) denote
the standard basis for ℂ𝑘 and (𝑒𝑗 ) the associated dual basis, the matrix entries of 𝜓𝛼

(𝜓𝛼 )𝑙𝑗 (𝑝) = 𝑒𝑙 (𝜎𝛼 (𝑝, 𝑒𝑗 )),

satisfy

so they are holomorphic by composition. To prove (3.3), we compute

(𝑝, 𝜏𝛼𝛽 (𝑝)𝑣) = Φ𝛼 ∘ Φ𝛽 −1 (𝑝, 𝑣)
−1
= (Φ′𝛼 ∘ 𝐹 ∘ Φ−1
𝛼 ) ∘ (Φ′𝛼 ∘ Φ′𝛽 −1 ) ∘ (Φ′𝛽 ∘ 𝐹 ∘ Φ−1
𝛽 )(𝑝, 𝑣)

= (𝑝, 𝜓𝛼 (𝑝)−1 𝜏𝛼𝛽

′
(𝑝)𝜓𝛽 (𝑝)𝑣). □

The next corollary shows that the bundle produced by the vector bundle con-
struction theorem is unique up to isomorphism.
Holomorphic Bundle Tools 75

Corollary 3.8. Suppose 𝐸 → 𝑀 and 𝐸 ′ → 𝑀 are holomorphic vector bundles of

rank 𝑘 over the same base, and both admit local trivializations over the same open
cover with the same transition functions. Then 𝐸 ≅ 𝐸 ′ .

Proof. Just apply Proposition 3.7 with 𝜓𝛼 ≡ Id for every 𝛼. □

Corollary 3.9. Suppose 𝐸 → 𝑀 is a holomorphic rank-𝑘 vector bundle, and 𝒰 =

{𝑈𝛼 }𝛼∈𝐴 is a trivializing cover for 𝐸 with transition functions 𝜏𝛼𝛽 . Then 𝐸 is trivial
if and only if for each 𝛼 ∈ 𝐴 there is a holomorphic map 𝜓𝛼 ∶ 𝑈𝛼 → GL(𝑘, ℂ) such
that 𝜏𝛼𝛽 = 𝜓𝛼−1 𝜓𝛽 |𝑈𝛼 ∩𝑈𝛽 for each 𝛼 and 𝛽.

Proof. This is the special case of Proposition 3.7 when each transition function 𝜏𝛼𝛽
′

is the identity matrix. □

vector bundles. Suppose 𝜋 ∶ 𝐸 → 𝑀 is a rank-𝑘 holomorphic vector bundle, 𝑁 is

We need to introduce one more important tool for constructing holomorphic

a complex manifold, and 𝑓 ∶ 𝑁 → 𝑀 is a holomorphic map. We define a vector

bundle 𝑓 ∗ 𝐸 → 𝑁, called the pullback of 𝑬 by 𝒇 , as follows. The total space of
𝑓 ∗ 𝐸 is the following subset of 𝑁 × 𝐸, called the fiber product of 𝑵 with 𝑬 over
𝑴:
𝑓 ∗ 𝐸 = {(𝑝, 𝑣) ∈ 𝑁 × 𝐸 ∶ 𝜋(𝑣) = 𝑓 (𝑝)}.
The projection 𝑓 ∗ 𝐸 → 𝑁 is just the restriction of the projection on the first factor:
(𝑝, 𝑣) ↦ 𝑝. The fiber of 𝑓 ∗ 𝐸 over a point 𝑝 ∈ 𝑁 is then {𝑝} × 𝐸𝑓 (𝑝) , which inherits
a complex vector space structure from that of 𝐸𝑓 (𝑝) . To make it into a holomorphic

Given 𝑝 ∈ 𝑁, choose a holomorphic local frame (𝑏1 , … , 𝑏𝑘 ) for 𝐸 on a neigh-

vector bundle, we construct local frames and use the chart lemma.

borhood of 𝑓 (𝑝), and define a rough local frame (𝑏̃1 , … , 𝑏̃𝑘 ) for 𝑓 ∗ 𝐸 on a neigh-
borhood of 𝑝 by
𝑏̃𝑗 (𝑝) = (𝑝, 𝑏𝑗 (𝑓 (𝑝))).
Where two such frames (𝑏𝑗 ) and (𝑏′𝑗 ) overlap, there is a holomorphic GL(𝑘, ℂ)-
valued transition function 𝜏 such that 𝑏′𝑗 (𝑞) = 𝜏𝑗𝑘 (𝑞)𝑏𝑘 (𝑞), and therefore

𝑏̃′𝑗 (𝑝) = (𝑝, 𝜏𝑗𝑘 (𝑓 (𝑝))𝑏̃𝑘 (𝑝)).

In other words, the frames (𝑏̃𝑗 ) and (𝑏′̃𝑗 ) overlap with the transition function 𝜏 ∘ 𝑓 ,
which is holomorphic by composition. It follows from the chart lemma that 𝑓 ∗ 𝐸
is a holomorphic vector bundle.

sections, defined as follows. For 𝜎 ∈ 𝒪(𝑈 ; 𝐸) over an open subset 𝑈 ⊆ 𝑀, we

In the situation described above, there is a pullback operator on local or global

define 𝑓 ∗ 𝜎 ∈ 𝒪(𝑓 −1 (𝑈 ); 𝑓 ∗ 𝐸) by
𝑓 ∗ 𝜎(𝑥) = (𝑥, 𝜎(𝑓 (𝑥))).
76 3. Holomorphic Vector Bundles

Proposition 3.10 (Properties of Pullbacks of Sections). Let 𝐸 → 𝑀 be a holo-

morphic vector bundle and 𝑓 ∶ 𝑁 → 𝑀 a holomorphic map.
(a) For each open set 𝑈 ⊆ 𝑁, pullback defines a complex-linear map
𝑓 ∗ ∶ 𝒪(𝑈 ; 𝐸) → 𝒪(𝑓 −1 (𝑈 ); 𝑓 ∗ 𝐸).
(b) If (𝑠1 , … , 𝑠𝑘 ) is a holomorphic local frame for 𝐸 over 𝑈 ⊆ 𝑀, then
(𝑓 ∗ 𝑠1 , … , 𝑓 ∗ 𝑠𝑘 ) is a holomorphic local frame for 𝑓 ∗ 𝐸 over 𝑓 −1 (𝑈 ).
(c) If 𝑓 is surjective, then 𝑓 ∗ ∶ 𝒪(𝑈 ; 𝐸) → 𝒪(𝑓 −1 (𝑈 ); 𝑓 ∗ 𝐸) is injective for
each 𝑈 .

► Exercise 3.11. Prove the preceding proposition.

Given a holomorphic vector bundle 𝜋 ∶ 𝐸 → 𝑀, a subset 𝐷 ⊆ 𝐸 is called

a holomorphic subbundle of 𝑬 if it is an embedded complex submanifold with
the property that each fiber 𝐷𝑝 = 𝐷 ∩ 𝐸𝑝 is a complex-linear subspace of some
fixed dimension 𝑚, and the restricted projection 𝜋|𝐷 ∶ 𝐷 → 𝑀 turns 𝐷 into a
holomorphic vector bundle over 𝑀.
Lemma 3.12 (Local Frame Criterion for Subbundles). Let 𝜋 ∶ 𝐸 → 𝑀 be a
holomorphic vector bundle, and let 𝐷 ⊆ 𝐸 be a subset whose intersection with
each fiber of 𝐸 is an 𝑚-dimensional complex subspace. Then 𝐷 is a holomorphic
subbundle if and only if each 𝑝 ∈ 𝑀 has a neighborhood over which there are
holomorphic sections 𝜎1 , … , 𝜎𝑚 of 𝐸 such that (𝜎1 (𝑞), … , 𝜎𝑚 (𝑞)) forms a basis for
𝐷𝑞 for each 𝑞 ∈ 𝑀.

Proof. Just like the proof of its smooth counterpart [LeeSM, Lemma 10.32]. □

Recall that a Hermitian inner product on a complex vector space 𝑉 is a map

from 𝑉 ×𝑉 to ℂ, often denoted by (𝑣, 𝑤) ↦ ⟨𝑣, 𝑤⟩, that has the following properties
for all 𝑣, 𝑤 ∈ 𝑉 and 𝑎 ∈ ℂ:
• CONJUGATE SYMMETRY: ⟨𝑣, 𝑤⟩ = ⟨𝑤, 𝑣⟩,
• SESQUILINEARITY: ⟨𝑎𝑣, 𝑤⟩ = 𝑎⟨𝑣, 𝑤⟩ and ⟨𝑣, 𝑎𝑤⟩ = 𝑎⟨𝑣, 𝑤⟩,
• POSITIVE DEFINITENESS: ⟨𝑣, 𝑣⟩ > 0 unless 𝑣 = 0.
Suppose 𝑀 is a smooth manifold and 𝐸 → 𝑀 is a smooth complex vector
bundle of rank 𝑚. A Hermitian fiber metric ℎ on 𝐸 is a choice ℎ𝑝 of Hermitian
inner product on each fiber 𝐸𝑝 that is smoothly varying in the sense that for any pair
of smooth sections 𝜎, 𝜏 of 𝐸 over an open subset 𝑈 ⊆ 𝑀, the value ℎ𝑝 (𝜎(𝑝), 𝜏(𝑝))
depends smoothly on 𝑝 ∈ 𝑈 . When it will not cause confusion, we will often denote
a Hermitian fiber metric by ⟨𝑣, 𝑤⟩ ∈ ℂ for 𝑣, 𝑤 ∈ 𝐸𝑝 , or ⟨𝜎, 𝜏⟩ ∈ 𝐶 ∞ (𝑀; ℂ)
for 𝜎, 𝜏 ∈ Γ(𝐸). Given a Hermitian fiber metric on 𝐸, we define the norm of a
vector 𝑣 ∈ 𝐸𝑝 as |𝑣| = ⟨𝑣, 𝑣⟩1/2 . A smooth complex vector bundle endowed with a
Hermitian fiber metric is called a Hermitian vector bundle. An easy partition-of-
unity argument shows that every smooth complex vector bundle admits a Hermitian
Holomorphic Bundle Tools 77

fiber metric. (Note that if 𝐸 is a holomorphic bundle, it does not make sense to ask
that a fiber metric be holomorphic, because of the conjugate linearity in the second
argument. Hermitian fiber metrics on holomorphic bundles are merely required to
be smooth.)

The Space of Holomorphic Sections

If 𝐸 → 𝑀 is a holomorphic vector bundle, recall that 𝒪(𝑀; 𝐸) denotes the
complex vector space of global holomorphic sections of 𝐸. The next theorem illus-
trates one of the most dramatic differences between smooth bundles and holomor-
phic ones.
Theorem 3.13. Suppose 𝑀 is a compact complex manifold and 𝐸 → 𝑀 is a
holomorphic vector bundle over 𝑀. Then 𝒪(𝑀; 𝐸) is finite-dimensional.
We present here a proof using complex analysis. A different proof based on the
theory of partial differential equations will be presented in Chapter 9 (just following
Theorem 9.35).
This proof will be based on the following standard lemma, which is proved in
most functional analysis texts. For convenience, we include a proof here.
Lemma 3.14. In a real or complex normed linear space 𝒳 , the closed unit ball is
compact if and only if 𝒳 is finite-dimensional.

Proof. Because all norms on a finite-dimensional vector space are equivalent, if 𝒳

gous result for ℝ𝑛 or ℂ𝑛 with its standard norm.

is finite-dimensional, compactness of the closed unit ball follows from the analo-

Conversely, suppose 𝒳 is infinite-dimensional. We will construct by induction

an infinite sequence {𝑥𝑗 }∞
𝑗=1 in 𝒳 with ‖𝑥𝑗 ‖ = 1 and ‖𝑥𝑗 − 𝑥𝑘 ‖ > 2 for all 𝑗 ≠ 𝑘;
1

closed unit ball in 𝒳 is not compact.

because such a subsequence can have no convergent subsequence, it shows that the

To begin the induction, just choose 𝑥1 ∈ 𝒳 arbitrarily with ‖𝑥1 ‖ = 1. As-

suming 𝑥1 , … , 𝑥𝑛 have been chosen, let 𝒳𝑛 = span(𝑥1 , … , 𝑥𝑛 ). Because 𝒳 is
infinite-dimensional, there is some 𝑦 ∈ 𝒳 ∖ 𝒳𝑛 . Since every finite-dimensional
subspace of a normed linear space is a closed subset, there is some 𝜀 > 0 such that
the open ball 𝐵𝜀 (𝑦) is contained in 𝒳 ∖ 𝒳𝑛 . Let 𝜀1 be the supremum of such 𝜀’s.
This means ‖𝑦 − 𝑥‖ ≥ 𝜀1 for all 𝑥 ∈ 𝒳𝑛 , and there is some 𝑥∗ ∈ 𝒳𝑛 such that
‖𝑦 − 𝑥∗ ‖ < 2𝜀1 . Set 𝑥𝑛+1 = (𝑦 − 𝑥∗ )/‖𝑦 − 𝑥∗ ‖, so that ‖𝑥𝑛+1 ‖ = 1 (see Fig. 3.1).
To show that 𝑥𝑛+1 satisfies the required conditions, suppose 𝑗 ∈ {1, … , 𝑛} and
compute

𝑦 − 𝑥∗ ‖𝑦 − 𝑥∗ ‖ 𝑥𝑗 ‖𝑦 − (𝑥 + ‖𝑦 − 𝑥 ‖ 𝑥𝑗 )‖
∗ ∗
‖𝑥𝑛+1 − 𝑥𝑗 ‖ = − = .
‖ ‖𝑦 − 𝑥∗ ‖ ‖𝑦 − 𝑥∗ ‖ ‖ ‖𝑦 − 𝑥∗ ‖
78 3. Holomorphic Vector Bundles

𝒳 𝑦 𝜀1 𝒳𝑛

𝑥∗

𝑥𝑛+1
0

Figure 3.1. Proof of Lemma 3.14

In the last expression on the right, the numerator is greater than or equal to 𝜀1
because (𝑥∗ + ‖𝑦 − 𝑥∗ ‖ 𝑥𝑗 ) ∈ 𝒳𝑛 ; and the denominator is less than 2𝜀1 by our
choice of 𝑥∗ . Therefore, ‖𝑥𝑛+1 − 𝑥𝑗 ‖ > 12 , thus completing the induction. □

Proof of Theorem 3.13. Choose a Hermitian fiber metric ⟨⋅, ⋅⟩ℎ on 𝐸, and use it
to define a global norm on 𝒪(𝑀; 𝐸) by

‖𝜎‖ = sup |𝜎(𝑝)|ℎ .

𝑝∈𝑀

norm is compact. Let {𝜎𝑘 }∞

𝑘=1 be any sequence in 𝒪(𝑀; 𝐸) with ‖𝜎𝑘 ‖ ≤ 1.
We will prove the theorem by showing that the closed unit ball with respect to this

For each 𝑝 ∈ 𝑀, there is a neighborhood 𝑊𝑝 of 𝑝 on which 𝐸 admits a holo-

morphic local frame. For each 𝑝, choose open sets 𝑈𝑝 , 𝑉𝑝 such that 𝑝 ∈ 𝑈𝑝 ⊆ 𝑈 𝑝 ⊆
𝑉𝑝 ⊆ 𝑉 𝑝 ⊆ 𝑊𝑝 . The collection of all such 𝑈𝑝 is an open cover of 𝑀, so we can
choose finitely many sets 𝑊1 , … , 𝑊𝑚 such that the corresponding sets 𝑈1 , … , 𝑈𝑚
cover 𝑀.
Consider the first of these sets 𝑊1 with its holomorphic local frame (𝑠1 , … , 𝑠𝑘 ).
On the restriction of 𝐸 to 𝑉 1 we have two fiber norms: the given one | ⋅ |ℎ , and the
“Euclidean” one determined by the local frame, namely

|𝜎(𝑝)|𝑒 = |𝜎 𝑗 (𝑝)𝑠𝑗 (𝑝)|𝑒 = (|𝜎 1 (𝑝)|2 + ⋯ + |𝜎 𝑘 (𝑝)|2 ) .

1/2

Let 𝑆 be the compact set 𝑉 1 × 𝕊2𝑘−1 ⊆ 𝑉 1 × ℂ𝑘 . The function

(𝑝, (𝑧1 , … , 𝑧𝑘 )) ↦ |𝑧𝑗 𝑠𝑗 (𝑝)|ℎ

and thus attains maximum and minimum positive values Λ and 𝜆 there.
determined by the given fiber metric is continuous and positive on this compact set,
Holomorphic Bundle Tools 79

For any 𝑝 ∈ 𝑉1 , if 𝜎(𝑝) ≠ 0, then the components of 𝜎(𝑝)/|𝜎(𝑝)|𝑒 lie in 𝕊2𝑛−1 ,

𝜎(𝑝)
so
|𝜎(𝑝)|ℎ = |𝜎(𝑝)|𝑒 ≥ |𝜎(𝑝)|𝑒 𝜆,
| |𝜎(𝑝)|𝑒 |ℎ
and a similar computation shows that |𝜎(𝑝)|ℎ ≤ Λ|𝜎(𝑝)|𝑒 . Of course, the same
inequalities hold trivially when 𝜎(𝑝) = 0.
Returning to our sequence {𝜎𝑘 } with ‖𝜎𝑘 ‖ ≤ 1, we conclude from the estimate
above that |𝜎𝑘 (𝑝)|𝑒 ≤ (1/𝜆) for 𝑝 ∈ 𝑉1 , and thus each component function of 𝜎𝑘
with respect to the given frame is uniformly bounded on 𝑉1 . Montel’s theorem

set 𝑈 1 ⊆ 𝑉1 to a section that is holomorphic there.

(Prop. 1.35) then implies that a subsequence converges uniformly on the compact

𝑈 2 , and so on for 𝑈 3 , … , 𝑈 𝑚 . The final subsequence converges in the global ‖ ⋅ ‖ℎ

Proceeding similarly, we can choose a sub-subsequence that also converges on

norm to a global holomorphic section, showing that the unit ball in 𝒪(𝑀; 𝐸) is
compact. □

Examples of Holomorphic Vector Bundles

Every complex manifold plays host to one important holomorphic vector bun-
dle, from which many others can be constructed.
Example 3.15 (The Holomorphic Tangent Bundle). Let 𝑀 be a complex 𝑛-
manifold and 𝑇 ′ 𝑀 be its holomorphic tangent bundle. Each local holomorphic
coordinate chart (𝑧𝑗 ) provides a local frame (𝜕/𝜕𝑧1 , … , 𝜕/𝜕𝑧𝑛 ) for 𝑇 ′ 𝑀. To see
how two such frames overlap, let (𝑤𝑗 ) be another local holomorphic coordinate
chart. The chain rule (Prop. 1.47) gives
𝜕 𝜕𝑧𝑗 𝜕 𝜕𝑧𝑗 𝜕
= + .
𝜕𝑤𝑘 𝜕𝑤𝑘 𝜕𝑧𝑗 𝜕𝑤𝑘 𝜕𝑧𝑗
Because each 𝑧𝑗 is holomorphic, the second term on the right is identically zero, so
the transition function from the frame (𝜕/𝜕𝑧𝑗 ) to the frame (𝜕/𝜕𝑤𝑘 ) is the matrix-
valued function (𝜕𝑧𝑗 /𝜕𝑤𝑘 ), which is holomorphic. Thus the chart lemma gives
𝑇 ′ 𝑀 the structure of a holomorphic vector bundle. //
Example 3.16 (Dual Bundles). Now suppose 𝐸 → 𝑀 is an arbitrary holomorphic
vector bundle. The dual bundle of 𝑬 is the bundle 𝐸 ∗ whose fiber at each point
𝑝 ∈ 𝑀 is the space 𝐸𝑝∗ of complex-linear functionals 𝐸𝑝 → ℂ. Given a local
holomorphic frame (𝑠1 , … , 𝑠𝑚 ) for 𝐸, we can form the dual frame (𝜀1 , … , 𝜀𝑚 ) for
𝐸 ∗ , satisfying 𝜀𝑗 (𝑝)(𝑠𝑘 (𝑝)) = 𝛿𝑗𝑘 for all 𝑝. If (𝑠1̃ , … , 𝑠𝑚̃ ) is another holomorphic
local frame for 𝐸 with transition function 𝜏 satisfying
𝑠𝑗̃ = 𝜏𝑗𝑘 𝑠𝑘 ,
then a little linear algebra shows that
𝜀𝑘̃ = (𝜏 −1 )𝑘𝑙 𝜀𝑙 ,
80 3. Holomorphic Vector Bundles

so the chart lemma shows that 𝐸 ∗ is a holomorphic vector bundle whose transition
functions are the transposed inverses (called the contragredients) of those of 𝐸. //
Example 3.17 (Bundles Over 0-Manifolds). If 𝑀 is a 0-manifold and 𝐸 → 𝑀 is
any complex vector bundle, for each 𝑝 ∈ 𝑀 we can choose a basis (𝑠1 (𝑝), … , 𝑠𝑘 (𝑝))
for the fiber 𝐸𝑝 , and the maps 𝑝 ↦ 𝑠𝑗 (𝑝) are vacuously holomorphic. Thus
(𝑠1 , … , 𝑠𝑘 ) is a holomorphic global frame, so every bundle over 𝑀 is holomor-
phically trivial. //
Example 3.18 (Whitney Sums). If 𝐸 → 𝑀 and 𝐸 ′ → 𝑀 are holomorphic vector
bundles of ranks 𝑘 and 𝑘′ , respectively, their Whitney sum is the bundle 𝐸 ⊕ 𝐸 ′
whose fiber at each 𝑝 ∈ 𝑀 is the direct sum 𝐸𝑝 ⊕ 𝐸𝑝′ . Given holomorphic local
frames (𝑠1 , … , 𝑠𝑘 ) for 𝐸 and (𝑠′1 , … , 𝑠′𝑘′ ) for 𝐸 ′ , we get a local frame (𝑠1 , … , 𝑠𝑘 ,
𝑠′1 , … , 𝑠′𝑘′ ) for 𝐸 ⊕ 𝐸 ′ . If 𝜏 and 𝜏 ′ are transition functions for overlapping local
frames for 𝐸 and 𝐸 ′ , respectively, then the transition function for 𝐸 ⊕ 𝐸 ′ is the
GL(𝑘 + 𝑘′ , ℂ)-valued matrix function ( 𝜏0 𝜏0′ ), which is holomorphic. Thus by the
chart lemma, 𝐸 ⊕ 𝐸 ′ is a holomorphic vector bundle of rank 𝑘 + 𝑘′ . //
Example 3.19 (Tensor Product Bundles). With 𝐸, 𝐸 ′ as in the previous example,
we define the tensor product bundle 𝐸 ⊗ 𝐸 ′ similarly, as the bundle whose fiber
at 𝑝 is 𝐸𝑝 ⊗ 𝐸𝑝′ . Any section of 𝐸 ⊗ 𝐸 ′ can be written locally as a finite sum
∑𝑗 𝜎𝑗 ⊗ 𝜎𝑗′ , where each 𝜎𝑗 is a local section of 𝐸 and each 𝜎𝑗′ is a local section of
𝐸 ′ . Local frames (𝑠𝑖 ) for 𝐸 and (𝑠′𝑗 ) for 𝐸 ′ yield a local frame (𝑠𝑖 ⊗ 𝑠′𝑗 ) for 𝐸 ⊗ 𝐸 ′ ,
with holomorphic transition functions, so 𝐸 ⊗ 𝐸 ′ is a holomorphic vector bundle
of rank 𝑘𝑘′ . //
Example 3.20 (Determinant Bundles). Suppose 𝐸 → 𝑀 is a holomorphic vec-
tor bundle of rank 𝑘. The determinant bundle of 𝑬 is the complex line bundle
det 𝐸 → 𝑀 whose fiber at a point 𝑝 ∈ 𝑀 is the 1-dimensional vector space Λ𝑘 (𝐸𝑝 )
of alternating 𝑘-vectors (i.e., contravariant alternating 𝑘-tensors) in 𝐸𝑝 . Any holo-
morphic local frame (𝑠1 , … , 𝑠𝑘 ) for 𝐸 yields a local frame 𝑠1 ∧ ⋯ ∧ 𝑠𝑘 for det 𝐸. If
𝜏 is a transition function between two local trivializations of 𝐸, then det(𝜏) is the
corresponding transition function for det 𝐸, so the determinant bundle is holomor-
phic. //
Example 3.21 (Restriction of a Bundle). If 𝜋 ∶ 𝐸 → 𝑀 is a holomorphic vec-
tor bundle and 𝑆 ⊆ 𝑀 is a complex submanifold, the set 𝐸|𝑆 = 𝜋 −1 (𝑆) together
with the projection inherited from 𝐸 is called the restriction of 𝑬 to 𝑺. Each holo-
moorphic local trivialization for 𝐸 restricts to a local trivialization for 𝐸|𝑆 , with
transition functions that are restrictions of those of 𝐸, so 𝐸|𝑆 is a holomorphic
vector bundle over 𝑆. //
Example 3.22 (Quotient Bundles). Suppose 𝐸 → 𝑀 is a holomorphic vector
bundle of rank 𝑛 and 𝐹 ⊆ 𝐸 is a holomorphic rank-𝑘 subbundle. The quotient
bundle 𝐸/𝐹 → 𝑀 is the bundle whose fiber at 𝑝 ∈ 𝑀 is the quotient space 𝐸𝑝 /𝐹𝑝 .
Holomorphic Line Bundles 81

Given a holomorphic local frame (𝑠1 , … , 𝑠𝑘 ) for 𝐹 , we can complete it to a holo-

morphic local frame (𝑠1 , … , 𝑠𝑛 ) for 𝐸 (perhaps after shrinking the domain), and
then consider (𝑠𝑘+1 , … , 𝑠𝑛 ) (mod 𝐹 ) as a local frame for 𝐸/𝐹 . Where two such
frames (𝑠𝑖 ) and (𝑠𝑗̃ ) overlap, the holomorphic transition matrix has the form

𝛼𝑗𝑖 0
𝜏𝑗𝑖 = ,
(𝛽𝑗𝑖 𝛾𝑗𝑖 )

and we have 𝑠𝑗̃ = ∑𝑛𝑖=𝑘+1 𝛾𝑗𝑖 𝑠𝑖 (mod 𝐹 ) for 𝑗 = 𝑘 + 1, … , 𝑛, so 𝐸/𝐹 is a holomor-

phic bundle of rank 𝑛 − 𝑘 by the chart lemma. //
Example 3.23 (Normal Bundles). Suppose 𝑀 is a complex 𝑛-manifold and 𝑆 ⊆
𝑀 is 𝑘-dimensional complex submanifold. The holomorphic normal bundle of
𝑺 in 𝑴 is the bundle 𝑁𝑆 → 𝑆 defined by 𝑁𝑆 = (𝑇 ′ 𝑀|𝑆 )/𝑇 ′ 𝑆. The result
of Example 3.22 shows that it is a holomorphic bundle of rank 𝑛 − 𝑘 on 𝑆. (It is

set of tangent vectors that are orthogonal to 𝑆 with respect to some Riemannian
important to observe that the geometric normal bundle that can be defined as the

metric on 𝑀 will not in general have holomorphic transition functions.) //

Example 3.24 (Homomorphism and Endomorphism Bundles). Again let 𝐸, 𝐸 ′

be holomorphic vector bundles over 𝑀. We can form the bundle Hom(𝐸, 𝐸 ′ )
whose fiber at a point 𝑝 ∈ 𝑀 is the space of complex-linear maps from 𝐸𝑝 to
𝐸𝑝′ . There is a canonical bijection from 𝐸 ′ ⊗ 𝐸 ∗ to Hom(𝐸, 𝐸 ′ ) which sends an
element 𝜎𝑗′ ⊗ 𝜑𝑗 ∈ 𝐸𝑥′ ⊗ 𝐸𝑥∗ to the linear map from 𝐸𝑥 to 𝐸𝑥′ given by 𝑒 ↦ 𝜑𝑗 (𝑒)𝜎𝑗′ .
Therefore, Hom(𝐸, 𝐸 ′ ) has local trivializations with the same transition functions
as 𝐸 ′ ⊗ 𝐸 ∗ , and is thus a holomorphic vector bundle canonically isomorphic to
𝐸 ′ ⊗ 𝐸 ∗ . In the special case 𝐸 = 𝐸 ′ , the bundle Hom(𝐸, 𝐸) is denoted by End(𝐸)
and called the endomorphism bundle of 𝑬. //

Holomorphic Line Bundles

morphic vector bundles of rank 1). Local trivializations of line bundles are most
For the remainder of the chapter, we focus on holomorphic line bundles (i.e., holo-

local sections; and transition functions are GL(1, ℂ)-valued holomorphic functions,
easily expressed in terms of local frames, which in this case are just nonvanishing

which we can consider as nonvanishing holomorphic scalar-valued functions. The

next lemma shows how transition functions are related to local frames.
Lemma 3.25. Let 𝐿 → 𝑀 be a holomorphic line bundle. Suppose (𝑈𝛼 , Φ𝛼 )
and (𝑈𝛽 , Φ𝛽 ) are holomorphic local trivializations of 𝐿, with transition function
𝜏𝛼𝛽 ∶ 𝑈𝛼 ∩ 𝑈𝛽 → GL(1, ℂ) from Φ𝛽 to Φ𝛼 . Then the holomorphic local frames 𝑠𝛼
and 𝑠𝛽 associated with these local trivializations are related on 𝑈𝛼 ∩ 𝑈𝛽 by
(3.4) 𝑠𝛽 = 𝜏𝛼𝛽 𝑠𝛼 .
82 3. Holomorphic Vector Bundles

Proof. By Lemma 3.3, for all 𝑝 ∈ 𝑈𝛼 ∩ 𝑈𝛽 we have

(3.5) 𝑠𝛼 (𝑝) = Φ−1
𝛼 (𝑝, 𝑒1 ), 𝑠𝛽 (𝑝) = Φ−1
𝛽 (𝑝, 𝑒1 ),

𝛽 (𝑝, 𝑒1 ) = (𝑝, 𝜏𝛼𝛽 (𝑝)𝑒1 ).

Φ𝛼 ∘ Φ−1
and by Lemma 3.1,

Therefore,

𝛽 (𝑝, 𝑒1 ) = (𝑝, 𝜏𝛼𝛽 (𝑝)𝑒1 ).

Φ𝛼 (𝑠𝛽 (𝑝)) = Φ𝛼 ∘ Φ−1
On the other hand, since Φ𝛼 is linear on fibers, (3.5) implies
Φ𝛼 (𝜏𝛼𝛽 (𝑝)𝑠𝛼 (𝑝)) = 𝜏𝛼𝛽 (𝑝)Φ𝛼 (𝑠𝛼 (𝑝)) = (𝑝, 𝜏𝛼𝛽 (𝑝)𝑒1 ).
Comparing the last two equations and using the fact that Φ𝛼 is injective proves the
result. □

To help remember (3.4), observe that while the transition function 𝜏𝛼𝛽 expresses
the transition from the Φ𝛽 trivialization to the Φ𝛼 trivialization (see (3.1)), when

the transition from the 𝛼 frame to the 𝛽 frame.

expressed in terms of frames, the same function goes the other way: it expresses

As a consequence, the transition functions of some related bundles can be ex-

pressed easily in terms of those of the original bundles.
Lemma 3.26. Suppose 𝐿 and 𝐿′ are holomorphic line bundles over 𝑀, and both

𝜏𝛼𝛽 and 𝜏𝛼𝛽

′
have local trivializations over the same trivializing cover with transition functions
, respectively.
(a) 𝐿∗ has transition functions 1/𝜏𝛼𝛽 .
(b) 𝐿 ⊗ 𝐿′ has transition functions 𝜏𝛼𝛽 𝜏𝛼𝛽
′
.
(c) 𝐿𝑘 = 𝐿 ⊗ ⋯ ⊗ 𝐿 has transition functions (𝜏𝛼𝛽 )𝑘 .

► Exercise 3.27. Prove this lemma.

Next we note some important properties of the tensor product operation on

holomorphic line bundles.
Lemma 3.28. Let 𝑀 be a complex manifold, and let 𝐿, 𝐿′ , 𝐿″ be holomorphic line
bundles over 𝑀.
(a) (𝐿 ⊗ 𝐿′ ) ⊗ 𝐿″ ≅ 𝐿 ⊗ (𝐿′ ⊗ 𝐿″ ).
(b) 𝐿 ⊗ 𝐿′ ≅ 𝐿′ ⊗ 𝐿.
(c) If 𝐿1 and 𝐿′1 are holomorphic line bundles isomorphic to 𝐿 and 𝐿′ , re-
spectively, then 𝐿1 ⊗ 𝐿′1 ≅ 𝐿 ⊗ 𝐿′ .
(d) 𝐿 ⊗ 𝐿∗ is a trivial bundle.
(e) If 𝐿′ is trivial, then 𝐿 ⊗ 𝐿′ ≅ 𝐿.
Holomorphic Line Bundles 83

Proof. These all follow from the formulas given in Lemma 3.26 together with the
result of Corollary 3.8. □

Thanks to these results, we can make the following definition. Let 𝑀 be a

morphic line bundles over 𝑀. (Although the class of all holomorphic line bundles
complex manifold, and let Pic(𝑀) denote the set of isomorphism classes of holo-

over 𝑀 is too big to be a set, the fact that isomorphism classes are determined by
their transition functions implies that the isomorphism classes do constitute a set.)

card group of 𝑴 after the French mathematician Charles Émile Picard, who first
The next theorem shows that it is an abelian group in a natural way, called the Pi-

defined it in the early twentieth century.

Theorem 3.29 (The Picard Group). If 𝑀 is a complex manifold, the product

operation defined by [𝐿1 ] ⋅ [𝐿2 ] = [𝐿1 ⊗ 𝐿2 ] turns the set Pic(𝑀) of isomorphism
classes of holomorphic line bundles over 𝑀 into an abelian group. The identity
element is the isomorphism class of any trivial bundle, and the inverse of [𝐿] is
[𝐿∗ ].

Proof. This follows easily from Lemma 3.28. □

The next lemma gives another way of understanding tensor power bundles.

Lemma 3.30. Let 𝐿 → 𝑀 be a holomorphic line bundle. For each positive integer
𝑑, the tensor power bundle (𝐿∗ )𝑑 is naturally isomorphic to the bundle whose fiber
at a point 𝑝 ∈ 𝑀 is the space of functions 𝜑 ∶ 𝐿𝑝 → ℂ that are homogeneous of
degree 𝒅, meaning that 𝜑(𝜆𝑣) = 𝜆𝑑 𝜑(𝑣) for all 𝜆 ∈ ℂ and 𝑣 ∈ 𝐿𝑝 . If 𝑠 ∶ 𝑈 → 𝐿
is a holomorphic local frame for 𝐿, then a local section 𝜑 of (𝐿∗ )𝑑 is holomorphic
on 𝑈 if and only if the function 𝑝 ↦ 𝜑(𝑠(𝑝)) is holomorphic.

Proof. Just as in smooth manifold theory, the fiber of (𝐿∗ )𝑑 over 𝑝 is naturally
isomorphic to the space of 𝑑-linear maps from 𝐿𝑝 × ⋯ × 𝐿𝑝 to ℂ (see [LeeSM,
Prop. 12.10]). Since 𝐿𝑝 is 1-dimensional, each such map 𝜑 gives rise to a homo-
geneous map 𝜑 ̃ ∶ 𝐿𝑝 → ℂ of degree 𝑑 by 𝜑 ̃ (𝑣) = 𝜑(𝑣, … , 𝑣); conversely, given
a homogeneous map 𝜑 ̃ , we can recover 𝜑 by choosing a basis vector 𝑏 for 𝐿𝑝 and
defining 𝜑(𝑏, … , 𝑏) = 𝜑
̃ (𝑏), and noting that a multilinear map is determined by its

If 𝑠 is a holomorphic local frame for 𝐿 and 𝜀 is the dual frame for 𝐿∗ , then
action on basis vectors.

𝜀 = 𝜀 ⊗ ⋯ ⊗ 𝜀 is a holomorphic local frame for (𝐿∗ )𝑑 whose action on 𝑠 as

𝑑

a homogeneous function is 𝜀𝑑 (𝑠) = 1. Therefore an arbitrary section 𝜑 of (𝐿∗ )𝑑

is a holomorphic multiple of this frame if and only if its action on 𝑠(𝑝) depends
holomorphically on 𝑝. □
84 3. Holomorphic Vector Bundles

Line Bundles over Projective Space

There is one particularly important line bundle over ℂℙ𝑛 . We define 𝑇 → ℂℙ𝑛 as
the following subbundle of the product bundle ℂℙ𝑛 × ℂ𝑛+1 :
𝑇 = {(𝜉, 𝑧) ∈ ℂℙ𝑛 × ℂ𝑛+1 ∶ 𝑧 ∈ 𝜉 }.
Proposition 3.31. 𝑇 → ℂℙ𝑛 is a holomorphic line bundle.

Proof. We will use the local frame criterion (Lemma 3.12). Because ℂℙ𝑛 ×ℂ𝑛+1 is

of ℂℙ𝑛 to ℂ𝑛+1 . On the open subset 𝑈𝛼 ⊆ ℂℙ𝑛 consisting of points [𝑤0 , … , 𝑤𝑛 ]

a product bundle, local sections of it are determined by functions from open subsets

with 𝑤𝛼 ≠ 0, we have a local section 𝑠𝛼 ∶ 𝑈𝛼 → 𝑇 defined by

1
𝑠𝛼 ([𝑤0 , … , 𝑤𝑛 ]) = ([𝑤], (𝑤0 , … , 𝑤𝑛 )).
𝑤𝛼
(3.6)

This local section is nowhere vanishing and spans the fiber of 𝑇 at each point of
𝑈𝛼 . In terms of affine coordinates (𝑧1 , … , 𝑧𝑛 ) ↔ [𝑧1 , … , 1, … , 𝑧𝑛 ], it has the local
representation (𝑧1 , … , 𝑧𝑛 ) ↦ (𝑧1 , … , 1, … , 𝑧𝑛 ), so it is holomorphic. The local
frame criterion shows that 𝑇 is a holomorphic line bundle. □

The bundle 𝑇 → ℂℙ𝑛 is called the tautological bundle because the fiber of 𝑇
over each 𝜉 ∈ ℂℙ𝑛 is exactly the line 𝜉 itself.
It will be useful to have explicit expressions for the transition functions for this
bundle and related ones.
Proposition 3.32 (Transition Functions on ℂℙ𝒏 ). On each open subset 𝑈𝛼 ⊆ ℂℙ𝑛
where 𝑤𝛼 ≠ 0, let 𝑠𝛼 ∶ 𝑈𝛼 → 𝑇 be the holomorphic section defined by (3.6). On
𝑈𝛼 ∩ 𝑈𝛽 , these sections satisfy 𝑠𝛽 = 𝜏𝛼𝛽 𝑠𝛼 , where
(3.7) 𝜏𝛼𝛽 ([𝑤]) = 𝑤𝛼 /𝑤𝛽 .
Thus 𝑇 ∗ , 𝑇 𝑘 and (𝑇 ∗ )𝑘 have trivializations over the same open sets, with transition
functions
(3.8) 𝜏𝛼𝛽 ([𝑤])𝑘 = (𝑤𝛼 /𝑤𝛽 )𝑘 for 𝑇 𝑘 ;
(3.9) 𝜏𝛼𝛽 ([𝑤])−1 = 𝑤𝛽 /𝑤𝛼 for 𝑇 ∗ ;
(3.10) 𝜏𝛼𝛽 ([𝑤])−𝑘 = (𝑤𝛽 /𝑤𝛼 )𝑘 for (𝑇 ∗ )𝑘 .

only calculate the transition functions for 𝑇 . Let 𝑈𝛽 ⊆ ℂℙ𝑛 be the set where 𝑤𝛽 ≠ 0
Proof. The formulas (3.8)–(3.10) follow from (3.7) and Lemma 3.26, so we need

and let 𝑠𝛽 ∶ 𝑈𝛽 → 𝑇 be the corresponding holomorphic section. Then (assuming

𝛼 < 𝛽 for simplicity), on 𝑈𝛼 ∩ 𝑈𝛽 we have
𝑤0 𝑤𝛼 𝑤𝛽 𝑤𝑛 𝑤𝛼
𝑠𝛽 ([𝑤]) = ([𝑤], ( , … , , … , , … , = 𝑠𝛼 ([𝑤]).
𝑤𝛽 𝑤𝛽 𝑤𝛽 𝑤𝛽 )) 𝑤𝛽
This proves (3.7). □
Line Bundles over Projective Space 85

Let us look more carefully at the structure of the tautological bundle. The first

ℂℙ𝑛 ×ℂ𝑛+1 , it is an embedded complex submanifold of dimension 𝑛+1 (the dimen-

thing to note is that because it is a holomorphic subbundle of the product bundle

we use the fact that in addition to the bundle projection 𝑇 → ℂℙ𝑛 , which is the re-
sion of the base plus the dimension of the fiber). To further understand its structure,

striction of the projection 𝜋1 ∶ ℂℙ𝑛 × ℂ𝑛+1 → ℂℙ𝑛 , we have another projection

from 𝑇 to ℂ𝑛+1 , namely the restriction of 𝜋2 ∶ ℂℙ𝑛 × ℂ𝑛+1 → ℂ𝑛+1 .

Proposition 3.33. Let 𝑇0 ⊆ 𝑇 be the image of the zero section of 𝑇 , and let
Π ∶ 𝑇 → ℂ𝑛+1 be the restriction of 𝜋2 ∶ ℂℙ𝑛 × ℂ𝑛+1 → ℂ𝑛+1 . Then Π−1 (0) = 𝑇0 ,
which is a complex submanifold of 𝑇 biholomorphic to ℂℙ𝑛 , and Π restricts to a
biholomorphism from 𝑇 ∖ 𝑇0 to ℂ𝑛+1 ∖ {0}.

Proof. Directly from the definition of Π, it follows that Π−1 (0) = {([𝑤], 0) ∶ [𝑤] ∈
ℂℙ𝑛 }, which is exactly the image of the zero section 𝜁 ∶ ℂℙ𝑛 → 𝑇 . Because the
bundle projection 𝜋 ∶ 𝑇 → ℂℙ𝑛 is a holomorphic left inverse for 𝜁, it follows that
𝜁 is a holomorphic embedding whose image is Π−1 (0).
On the other hand, the restriction Π|𝑇 ∖𝑇0 ∶ 𝑇 ∖ 𝑇0 → ℂ𝑛+1 ∖ {0} has a holomor-
phic inverse given by 𝑣 ↦ ([𝑣], 𝑣), so it is a biholomorphism. □

Blowing Up
Proposition 3.33 shows that if 𝑇 is the tautological bundle over ℂℙ𝑛−1 , we can
think of the total space of 𝑇 as a copy of ℂ𝑛 in which the origin has been replaced
by a copy of ℂℙ𝑛−1 . This leads to a new way of constructing complex manifolds
out of other complex manifolds, by replacing a point in an 𝑛-manifold with a copy
of ℂℙ𝑛−1 , modeled locally on the total space of 𝑇 .
Let 𝑀 be a complex 𝑛-manifold and 𝑝 ∈ 𝑀. The blowup of 𝑴 at 𝒑 is a
complex manifold 𝑀 ̃ together with a surjective holomorphic map 𝜋 ∶ 𝑀 ̃ → 𝑀,
̃
defined as follows. As a set, 𝑀 is the disjoint union of 𝑀 ∖ {𝑝} with ℙ(𝑇𝑝′ 𝑀)
(the projectivization of the holomorphic tangent space 𝑇𝑝′ 𝑀). Let 𝐸 = ℙ(𝑇𝑝′ 𝑀) ⊆
𝑀̃ (called the exceptional hypersurface of the blowup), and define 𝜋 ∶ 𝑀 ̃ →𝑀
(called the blowdown map) by

̃ ∖ 𝐸,
𝑧, 𝑧 ∈ 𝑀
𝜋(𝑧) =
{𝑝, 𝑧 ∈ 𝐸.

Let 𝑇 be the tautological bundle over ℂℙ𝑛−1 , and let Π ∶ 𝑇 → ℂ𝑛 be the restriction
to 𝑇 of 𝜋2 ∶ ℂℙ𝑛−1 × ℂ𝑛 → ℂ𝑛 , called the model blowdown map.
86 3. Holomorphic Vector Bundles

Given a holomorphic coordinate chart (𝑈 , 𝜑) for 𝑀 centered at 𝑝, with coordi-

nate functions (𝑧1 , … , 𝑧𝑛 ), choose 𝜀 > 0 such that 𝐵𝜀 (0) ⊆ 𝜑(𝑈 ). Let

𝑈 (𝜀) = 𝜑−1 (𝐵𝜀 (0)) ⊆ 𝑀,

̃ (𝜀) = 𝜋 −1 (𝑈 (𝜀)) ⊆ 𝑀,
̃
(3.11)
(3.12) 𝑈
(3.13) 𝑇 (𝜀) = Π−1 (𝐵𝜀 (0)) ⊆ 𝑇 .

̃ (𝜀) by
Define a map Φ ∶ 𝑇 (𝜀) → 𝑈

⎧𝜑 (𝑧), 𝑧 ≠ 0,
−1
⎪
Φ([𝑤], 𝑧) = ⎨ 1 𝜕 𝜕
⎪[𝑤 + ⋯ + 𝑤𝑛 𝑛 | ], 𝑧 = 0.
𝜕𝑧 |𝑝
(3.14)
⎩ 1 𝜕𝑧 𝑝

In the case when 𝑀 has complex dimension 1, ℙ(𝑇𝑝′ 𝑀) is a 0-dimensional

̃ =𝑀
projective space, which is just a single point; in that case we just define 𝑀
and 𝜋 to be the identity map.

Proposition 3.34 (Blowups Are Complex Manifolds). Let 𝑀 be a complex man-

̃ → 𝑀 the blowup of 𝑀 at a point 𝑝 ∈ 𝑀, and 𝐸 = 𝜋 −1 (𝑝). There is a
ifold, 𝜋 ∶ 𝑀
̃ with the property that for each holomor-
unique complex manifold structure on 𝑀
phic coordinate chart for 𝑀 centered at 𝑝, the map Φ defined above is a biholo-
morphism onto a neighborhood of 𝐸. The blowdown map 𝜋 is holomorphic, and 𝐸
is an embedded compact complex hypersurface.

Proof. We will construct coordinate charts for 𝑀 ̃ and use the complex manifold

Since the blowdown map 𝜋 restricts to a bijection from 𝑀 ̃ ∖𝐸 to 𝑀 ∖{𝑝}, we can

chart lemma to give it the structure of a complex manifold.

begin with all of the given holomorphic coordinate charts on 𝑀 ∖ {𝑝}, pulled back
to 𝑀̃ ∖𝐸 via 𝜋. To define holomorphic coordinates on a neighborhood of 𝐸, choose
a holomorphic coordinate chart (𝑈 , 𝜑) for 𝑀 centered at 𝑝 and define Φ ∶ 𝑇 (𝜀) →
𝑈̃ (𝜀) as above. This is a bijection, so we can use holomorphic coordinates on 𝑇 (𝜀)
composed with Φ−1 to define holomorphic coordinates on 𝑈 ̃ (𝜀). These are certainly
all holomorphically compatible with each other, and because the restriction of Φ to
𝑇 (𝜀) ∖ 𝑇0 is a biholomorphism onto 𝑈 ̃ (𝜀) ∖ 𝐸 ≈ 𝑈 (𝜀) ∖ {𝑝}, these coordinates are
also holomorphically compatible with the given coordinates on 𝑀 ̃ ∖ 𝐸. The only

chart (𝑈̃, 𝜑
̃ ) centered at 𝑝 are compatible with these.
remaining thing to check is that the coordinates defined by any other holomorphic

To that end, suppose (𝑈 ̃, 𝜑̃ ) is another chart centered at 𝑝, and let Φ̃ be the

corresponding map defined in the same way as Φ. Write the component functions
of 𝜑̃ ∘ 𝜑−1 (𝑧) as (𝑧1̃ (𝑧), … , 𝑧𝑛̃ (𝑧)). The composite map Φ ̃ −1 ∘ Φ has the following
Line Bundles over Projective Space 87

form:

⎧([𝑧1̃ (𝑧), … , 𝑧𝑛̃ (𝑧)], (𝑧1̃ (𝑧), … , 𝑧𝑛̃ (𝑧))), 𝑧 ≠ 0,

̃ −1 ∘ Φ([𝑤], 𝑧) = ⎪
Φ ⎨ 𝑗 𝜕 𝑧̃ 𝑗 𝜕 𝑧̃
1 𝑛
⎪([∑ 𝜕𝑧𝑗 𝑤 (0), … , 𝑤 (0) , 0 , 𝑧 = 0.
∑ 𝜕𝑧𝑗 ] )
⎩ 𝑗 𝑗

series for 𝑧𝑘̃ together with the fact that 𝑧𝑘̃ (0) = 0, we can write
To see that this is holomorphic, we will express it in another way. Using the Taylor

𝑛
𝑧𝑘̃ (𝑧) = 𝑧𝑗 𝑓𝑗𝑘 (𝑧)
∑
𝑗=1

for some holomorphic functions 𝑓𝑗𝑘 , with

𝜕 𝑧𝑘̃
𝑓𝑗𝑘 (0) = (0).
𝜕𝑧𝑗
Thus the composite map Φ̃ −1 ∘ Φ is the restriction to 𝑇 (𝜀) of the holomorphic map
from ℂℙ 𝑛−1
× ℂ to itself given by
𝑛

([𝑤], 𝑧) ↦ ([ 𝑤𝑗 𝑓𝑗1 (𝑧), … , 𝑤𝑗 𝑓𝑗𝑛 (𝑧)], (𝑧1̃ (𝑧), … , 𝑧𝑛̃ (𝑧))).

∑ ∑
𝑗 𝑗

This shows that Φ̃ −1 ∘ Φ is holomorphic, and the same argument applies to its in-
verse. Thus we have made 𝑀 ̃ into a complex manifold, and each of the maps Φ
defined above is a biholomorphism onto a neighborhood of 𝐸 as claimed. Since
any other such holomorphic structure must contain all of these charts, it must be

To see that 𝜋 is holomorphic, just note that its restriction to 𝑀 ̃ ∖ 𝐸 is a biholo-

equal to this one.

morphism onto its image, and its restriction to Φ(𝑇 (𝜀)) is equal to the holomorphic
composition 𝜑−1 ∘ Π ∘ Φ−1 . Because 𝐸 is the image of the compact complex hy-
persurface 𝑇0 under the diffeomorphism Φ ∶ 𝑇 (𝜀) → 𝑈 ̃ (𝜀), it is itself an embedded
compact complex hypersurface. □

Similarly, we can define the blowup of 𝑀 at finitely many points 𝑝1 , … , 𝑝𝑚 ∈

𝑀 by letting 𝑀̃ be the set

̃ = 𝑀 ∖ {𝑝1 , … , 𝑝𝑚 } ⨿ ℙ(𝑇𝑝′ 𝑀) ⨿ ⋯ ⨿ ℙ(𝑇𝑝′ 𝑀),

𝑀 1 𝑚

with the obvious blowdown map 𝜋 ∶ 𝑀 ̃ → 𝑀, and with the holomorphic struc-
̃
ture on 𝑀 defined by applying the above construction in a neighborhood of each
exceptional hypersurface 𝜋 −1 (𝑝𝑖 ).
This blowup construction will play a major role in our proof of the Kodaira
embedding theorem in Chapter 10.
88 3. Holomorphic Vector Bundles

Holomorphic Sections over Projective Space

bundles over ℂℙ𝑛 . We begin with the tautological bundle.

We can describe explicitly the spaces of holomorphic sections of various line

Proposition 3.35. The tautological bundle 𝑇 → ℂℙ𝑛 has no nontrivial global

holomorphic sections.

Proof. Suppose 𝜎 ∶ ℂℙ𝑛 → 𝑇 is a holomorphic section. Composing with the

model blowdown map Π ∶ 𝑇 → ℂ𝑛+1 , we get a holomorphic map Π ∘ 𝜎 ∶ ℂℙ𝑛 →
ℂ𝑛+1 . Because ℂℙ𝑛 is compact, each of the component functions of this map must
be constant. Thus there is some 𝑐 ∈ ℂ𝑛+1 such that Π ∘ 𝜎(ℂℙ𝑛 ) = {𝑐}.
For each 𝜉 ∈ ℂℙ𝑛 , the value 𝜎(𝜉) lies in the fiber of 𝑇 over 𝜉, and thus the point
𝑐 = Π(𝜎(𝜉)) lies in the 1-dimensional subspace 𝜉 ⊆ ℂ𝑛+1 . The only point in ℂ𝑛+1
that lies in every 1-dimensional subspace is the origin, and the preimage of 0 in 𝑇
contains only the zero point of each fiber, so 𝜎 is equal to the zero section. □

Next we look at the dual bundle of 𝑇 . Because it will turn out to be even more
important than 𝑇 itself, we use the symbol 𝐻 to denote the holomorphic line bundle
𝑇 ∗ ; the significance of this choice of notation will be explained in the next section.
Similarly, we use the notations
𝐻 0 for the trivial bundle ℂℙ𝑛 × ℂ → ℂℙ𝑛 ,
𝐻 𝑑 for the 𝑑-fold tensor power bundle 𝐻 ⊗ ⋯ ⊗ 𝐻 (𝑑 ≥ 1),
𝐻 −1 for 𝑇 ≅ 𝐻 ∗ ,
𝐻 −𝑑 for the 𝑑-fold tensor power bundle 𝑇 ⊗ ⋯ ⊗ 𝑇 (𝑑 ≥ 1).
The notation for negative powers of 𝐻 is motivated by the fact that 𝐻 −𝑑 = 𝑇 𝑑 rep-
resents the inverse of 𝐻 𝑑 in the Picard group. (In the algebraic geometry literature,
the notation 𝒪(𝑑) is often used to denote the line bundle 𝐻 𝑑 ; see the discussion

We can construct some nontrivial sections of 𝐻 and its positive tensor powers
following Exercise 5.15 for an explanation of the reason for this notation.)

in the following way. By Lemma 3.30, for 𝑑 ≥ 1, the fiber of 𝐻 𝑑 at 𝜉 ∈ ℂℙ𝑛 can
be identified with the set of functions from the line 𝜉 to ℂ that are homogeneous of
degree 𝑑. One way of constructing a section of 𝐻 𝑑 is to start with a homogeneous
holomorphic degree-𝑑 polynomial 𝑓 ∶ ℂ𝑛+1 → ℂ and restrict it to each 1-dimen-
sional subspace of ℂ𝑛+1 . This yields a rough section 𝜑𝑓 ∶ ℂℙ𝑛 → 𝐻 𝑑 , given
explicitly by
(3.15) 𝜑𝑓 (𝜉) = 𝑓 |𝜉 .
To see that this section is holomorphic, choose affine coordinates (𝑧1 , … , 𝑧𝑛 ) ↔
[𝑧1 , … , 1, … , 𝑧𝑛 ] on the open set 𝑈𝛼 ⊆ ℂℙ𝑛 where the 𝛼th homogeneous coordi-
nate is nonzero. Over 𝑈𝛼 , we have a local holomorphic frame 𝑠𝛼 ∶ 𝑈𝛼 → 𝑇 defined
Line Bundles over Projective Space 89

by (3.6). The section 𝜑𝑓 applied to 𝑠𝛼 gives

𝜑𝑓 (𝑠𝛼 ([𝑧1 , … , 1, … , 𝑧𝑛 ])) = 𝑓 (𝑧1 , … , 1, … , 𝑧𝑛 ),

which is holomorphic, so it follows from Lemma 3.30 that 𝜑𝑓 is a holomorphic

section.

phic sections of powers of 𝐻.

The next theorem shows that we have produced all of the nontrivial holomor-

Theorem 3.36 (Holomorphic Sections of Projective Line Bundles). Let 𝐻 →

ℂℙ𝑛 be the dual of the tautological bundle. The global holomorphic sections of
powers of 𝐻 are described as follows. Let 𝑑 be a positive integer.

𝒪(ℂℙ𝑛 ; 𝐻 −𝑑 ) = {0}.
𝒪(ℂℙ𝑛 ; 𝐻 0 ) = {constants}.
𝒪(ℂℙ𝑛 ; 𝐻 𝑑 ) = {𝜑𝑓 ∶ 𝑓 a homogeneous holomorphic degree-𝑑 polynomial}.

Proof. Proposition 3.35 showed that the only global holomorphic section of 𝑇 =
𝐻 −1 is the zero section. Assume for the sake of contradiction that 𝜎 is a nontrivial
holomorphic section of 𝐻 −𝑑 for some 𝑑 ≥ 2. We showed above that 𝐻 𝑑−1 has
a nontrivial holomorphic section 𝜑𝑓 . Then 𝜎 ⊗ 𝜑𝑓 is a nontrivial holomorphic
section of 𝐻 −𝑑 ⊗ 𝐻 𝑑−1 ≅ 𝐻 −1 , which is a contradiction.
Because a holomorphic section of the trivial bundle 𝐻 0 is just a scalar-valued

phic sections of 𝐻 0 are constants.

holomorphic function, it follows from Corollary 1.33 that the only global holomor-

Now consider 𝐻 𝑑 for 𝑑 > 0. We observed above that every homogeneous

holomorphic degree-𝑑 polynomial 𝑓 gives rise to a holomorphic section 𝜑𝑓 , so
we just need to show that every holomorphic section is of this form. Let 𝜎 ∈
𝒪(ℂℙ𝑛 ; 𝐻 𝑑 ) be arbitrary. We can define a holomorphic function 𝑓 ∶ ℂ𝑛+1 ∖ {0}
→ ℂ by
𝑓 (𝑧) = 𝜎([𝑧])(𝑧),
where we are viewing 𝜎([𝑧]) ∈ (𝐻[𝑧] )𝑑 as a homogeneous function from the line
[𝑧] to ℂ. By Hartogs’s extension theorem, 𝑓 extends to a holomorphic function on
all of ℂ𝑛+1 . It satisfies 𝑓 (𝜆𝑧) = 𝜆𝑑 𝑓 (𝑧) for all 𝜆 ∈ ℂ and 𝑧 ∈ ℂ𝑛+1 ∖ {0}, and thus
also for 𝑧 = 0 by continuity; in other words, it is homogeneous of degree 𝑑.
We will show that 𝑓 is actually a polynomial. For nonzero 𝑧 ∈ ℂ𝑛+1 , homo-

𝑧
geneity implies
|𝑓 (𝑧)| = |𝑧|𝑑 |𝑓 ( )| ≤ 𝐶|𝑧|𝑑 ,
|𝑧|
where 𝐶 is the supremum of |𝑓 | on the unit sphere. This implies that the Taylor
series of 𝑓 at the origin has no terms of order less than 𝑑. Let 𝑝 be the polynomial
90 3. Holomorphic Vector Bundles

function obtained by considering only the terms of degree 𝑑 in the Taylor series for
𝑓 at the origin:
𝑝(𝑧) = 𝑎𝑘1 …𝑘𝑛 (𝑧1 )𝑘1 ⋯ (𝑧𝑛 )𝑘𝑛 ,
∑
𝑘1 +⋯+𝑘𝑛 =𝑑
and let 𝑟(𝑧) = 𝑓 (𝑧) − 𝑝(𝑧). Then 𝑟 is homogeneous of degree 𝑑 because 𝑓 and 𝑝 are.
Since the Taylor series of 𝑟 starts with terms of order 𝑑 + 1, there is some constant
𝐶 ′ such that |𝑟(𝑧)| ≤ 𝐶 ′ |𝑧|𝑑+1 for all 𝑧 in the closed unit ball. For any 𝑧 ∈ ℂ𝑛+1
and 𝜀 > 0 small enough that |𝜀𝑧| ≤ 1, we have
|𝑟(𝑧)| = 𝜀−𝑑 |𝑟(𝜀𝑧)| ≤ 𝜀−𝑑 𝐶 ′ |𝜀𝑧|𝑑+1 = 𝜀𝐶 ′ |𝑧|.
Taking the limit as 𝜀 → 0, we find that 𝑟(𝑧) ≡ 0. Thus 𝑓 is equal to the polynomial
𝑝, and 𝜎 = 𝜑𝑝 . □

We will see later that every holomorphic line bundle on ℂℙ𝑛 is isomorphic to
𝐻 for some integer 𝑑 (see Proposition 9.51).
𝑑

Applications of Holomorphic Line Bundles

We end the chapter by describing two significant applications of holomorphic line
bundles.

Line Bundles and Hypersurfaces

In ℂ𝑛 , many complex hypersurfaces can be written as regular level sets of glob-
ally defined holomorphic functions. But in a compact complex manifold, this is
never possible, because all global holomorphic functions are constants. Instead,

Suppose 𝑀 is a complex manifold, 𝐿 → 𝑀 is a holomorphic line bundle, and

we can use sections of line bundles.

𝜎 ∈ 𝒪(𝑀; 𝐿). The variety determined by 𝝈 is the set 𝑉𝜎 = {𝑝 ∈ 𝑀 ∶ 𝜎(𝑝) = 0}.

We say the section 𝜎 vanishes simply if whenever 𝑠 is a local holomorphic frame
for 𝐿, we can write 𝜎 = 𝑓 𝑠 where 𝑑𝑓𝑝 ≠ 0 whenever 𝑓 (𝑝) = 0. It is easy to
check that this condition is independent of the choice of local frame, and if 𝜎 is a
holomorphic section that vanishes simply, then 𝑉𝜎 is a closed complex hypersurface
(i.e., a codimension-1 complex submanifold that is a closed subset of 𝑀).
Example 3.37. Suppose 𝑓 ∶ ℂ𝑛+1 → ℂ is a homogeneous holomorphic polyno-
mial of degree 𝑑, and 𝑉 ⊆ ℂℙ𝑛 is the algebraic variety determined by 𝑓 . The
polynomial 𝑓 also defines a holomorphic section 𝜑𝑓 ∶ ℂℙ𝑛 → 𝐻 𝑑 as in (3.15),
and 𝑉 is exactly the variety determined by this section.
In particular, in the case 𝑑 = 1, holomorphic sections of 𝐻 are of the form
𝜑𝑓 where 𝑓 is a complex-linear functional on ℂ𝑛+1 , and the varieties determined
by such sections are exactly the projective hyperplanes in ℂℙ𝑛 . For this reason,
the bundle 𝐻 is called the hyperplane bundle (which explains our choice of the
notation 𝐻). //
Applications of Holomorphic Line Bundles 91

section of a line bundle in this way. If 𝑀 is a complex manifold and 𝑆 ⊆ 𝑀 is a

It turns out that every closed complex hypersurface is cut out by a holomorphic

closed complex hypersurface, let us say an open cover {𝑈𝛼 }𝛼∈𝐴 of 𝑀 together with
holomorphic functions 𝑓𝛼 ∶ 𝑈𝛼 → ℂ is a system of local defining functions for 𝑺
if each function 𝑓𝛼 vanishes simply on 𝑈𝛼 ∩ 𝑆 and nowhere else. Every closed

defining function in a neighborhood of each point of 𝑆, and because 𝑆 is closed we

complex hypersurface has such a system: Corollary 2.13 shows that there is a local

can obtain an open cover of 𝑀 by adding in the open set 𝑈0 = 𝑀 ∖ 𝑆 with 𝑓0 ≡ 1.

The next lemma provides a weak sort of uniqueness for local defining functions.
Lemma 3.38. Suppose 𝑀 is a complex manifold and 𝑆 ⊆ 𝑀 is a complex hyper-
surface. If 𝑈 ⊆ 𝑀 is open and 𝑓1 , 𝑓2 ∶ 𝑈 → ℂ are holomorphic functions that
vanish simply on 𝑈 ∩𝑆 and nowhere else, then there is a nonvanishing holomorphic
function ℎ ∶ 𝑈 → ℂ such that 𝑓2 = ℎ𝑓1 .

Proof. Given 𝑓1 , 𝑓2 as in the statement of the lemma, for each point 𝑝 ∈ 𝑈 ∩ 𝑆

(if there are any), we can choose local holomorphic slice coordinates (𝑧1 , … , 𝑧𝑛 )
on a neighborhood 𝑊 of 𝑝 such that 𝑆 ∩ 𝑊 is the set where 𝑧𝑛 = 0. Since
𝑓1 (𝑧1 , … , 𝑧𝑛−1 , 0) ≡ 0, the Taylor series of 𝑓1 centered at any point of 𝑊 ∩ 𝑆
has no terms that are of order zero in the variable 𝑧𝑛 , and we can factor out 𝑧𝑛 to
write 𝑓1 (𝑧) = 𝑧𝑛 𝑔1 (𝑧) for some holomorphic function 𝑔1 . The fact that 𝑓1 vanishes
simply on 𝑈 ∩ 𝑆 implies that 𝑑𝑓1 is nonvanishing at points where 𝑧𝑛 = 0, so we
must have 𝑔1 ≠ 0 on on 𝑊 ∩ 𝑆. Similarly, 𝑓2 (𝑧) = 𝑧𝑛 𝑔2 (𝑧) with 𝑔2 nonzero on
𝑊 ∩ 𝑆. It follows that everywhere in 𝑊 ∖ 𝑆,
𝑓2 (𝑧) 𝑔2 (𝑧)
= ,
𝑓1 (𝑧) 𝑔1 (𝑧)
which extends to a nonvanishing holomorphic function on all of 𝑊 . Since the
complement of 𝑆 is dense in 𝑈 , the extension is unique by continuity. Thus 𝑓2 /𝑓1

each point of 𝑈 , and by uniqueness the extensions all fit together to determine a
extends uniquely to a nonvanishing holomorphic function in a neighborhood of

nowhere-vanishing holomorphic function ℎ ∶ 𝑈 → ℂ satisfying 𝑓2 = ℎ𝑓1 . □

Theorem 3.39 (Line Bundle Associated with a Hypersurface). Suppose 𝑀 is a
complex manifold and 𝑆 ⊆ 𝑀 is a closed complex hypersurface. Then there exist
a holomorphic line bundle 𝐿𝑆 → 𝑀, called the associated line bundle for 𝑺, and
a holomorphic section 𝜎 ∈ 𝒪(𝑀; 𝐿𝑆 ) that vanishes simply on 𝑆 and nowhere else.
Any two line bundles that admit holomorphic sections vanishing simply only on 𝑆
are isomorphic.

Proof. Let {(𝑈𝛼 , 𝑓𝛼 )}𝛼∈𝐴 be a system of local defining functions for 𝑆. Whenever
𝑈𝛼 ∩ 𝑈𝛽 ≠ ∅, Lemma 3.38 shows that that there is a nowhere-vanishing holomor-
phic function 𝜏𝛼𝛽 ∶ 𝑈𝛼 ∩ 𝑈𝛽 → GL(1, ℂ) such that 𝜏𝛼𝛽 = 𝑓𝛼 /𝑓𝛽 on the complement
of 𝑆.
92 3. Holomorphic Vector Bundles

When 𝑈𝛼 ∩ 𝑈𝛽 ∩ 𝑈𝛾 ≠ ∅, on the complement of 𝑆 we have

𝑓𝛼 𝑓𝛽
𝜏𝛼𝛽 𝜏𝛽𝛾 = = 𝜏𝛼𝛾 ,
𝑓𝛽 𝑓𝛾
and the same holds by continuity on all of 𝑈𝛼 ∩ 𝑈𝛽 ∩ 𝑈𝛾 . Thus we have the data

𝐿𝑆 → 𝑀, with a trivialization over each set 𝑈𝛼 and transition functions 𝜏𝛼𝛽 . The
to apply the bundle construction theorem and obtain a holomorphic line bundle

associated local frames 𝑠𝛼 ∶ 𝑈𝛼 → 𝐿𝑆 satisfy (3.4).

We define a holomorphic section 𝜎 ∶ 𝑀 → 𝐿𝑆 by setting
𝜎(𝑝) = 𝑓𝛼 (𝑝)𝑠𝛼 (𝑝) for 𝑝 ∈ 𝑈𝛼 .

follows immediately that 𝑆 is the variety determined by 𝜎, and 𝜎 vanishes simply

Equation (3.4) guarantees that these definitions agree where they overlap, and it

on 𝑆.
To show that any two such bundles are isomorphic, suppose 𝐿′ → 𝑀 is another
holomorphic line bundle and 𝜎 ′ ∶ 𝑀 → 𝐿′ is a holomorphic section that vanishes
simply on 𝑆 and nowhere else. We can choose an open cover {𝑈𝛼 }𝛼∈𝐴 that is
a trivializing cover for both bundles. Let 𝑠𝛼 ∶ 𝑈𝛼 → 𝐿𝑆 and 𝑠′𝛼 ∶ 𝑈𝛼 → 𝐿′ be
holomorphic local frames, and let 𝜏𝛼𝛽 and 𝜏𝛼𝛽
′
be the associated transition functions
so that
(3.16) 𝑠𝛽 = 𝜏𝛼𝛽 𝑠𝛼 , 𝑠′𝛽 = 𝜏𝛼𝛽
′ ′
𝑠𝛼 .

In each open set 𝑈𝛼 , we can write 𝜎 = 𝑓𝛼 𝑠𝛼 and 𝜎 ′ = 𝑓𝛼′ 𝑠′𝛼 for some holomor-
phic functions 𝑓𝛼 , 𝑓𝛼′ that vanish simply on 𝑆. Lemma 3.38 shows that 𝜓𝛼 = 𝑓𝛼′ /𝑓𝛼
extends to a nonvanishing holomorphic function on all of 𝑈𝛼 .
On 𝑈𝛼 ∩ 𝑈𝛽 , we have
𝑓𝛼 𝑠𝛼 = 𝜎 = 𝑓𝛽 𝑠𝛽 = 𝑓𝛽 𝜏𝛼𝛽 𝑠𝛼 ,
which implies 𝑓𝛼 = 𝑓𝛽 𝜏𝛼𝛽 , and similarly 𝑓𝛼′ = 𝑓𝛽′ 𝜏𝛼𝛽
′
. Therefore, on (𝑈𝛼 ∩ 𝑈𝛽 ) ∖ 𝑆,

(𝑓𝛼′ )(𝑓𝛽 𝜏𝛼𝛽 ) (𝑓𝛽 𝜏𝛼𝛽 )(𝑓𝛼 )

′ ′
we have

𝜏𝛼𝛽 = = = 𝜓𝛼−1 𝜏𝛼𝛽

′
𝜓𝛽 ,
𝑓𝛼′ 𝑓𝛽 𝑓𝛼′ 𝑓𝛽
and the same formula holds on all of 𝑈𝛼 ∩ 𝑈𝛽 by continuity. Thus 𝐿′ ≅ 𝐿𝑆 by the
isomorphism criterion (Prop. 3.7). □

pose 𝑆 ⊆ ℂℙ𝑛 is a nonsingular projective algebraic hypersurface defined by a sin-

Example 3.40 (Line Bundles Associated with Algebraic Hypersurfaces). Sup-

gle homogeneous polynomial of degree 𝑑. Example 3.37 showed there is a global

holomorphic section of 𝐻 𝑑 that vanishes simply on 𝑆 and nowhere else; thus by
uniqueness, 𝐿𝑆 ≅ 𝐻 𝑑 . //
Applications of Holomorphic Line Bundles 93

Line Bundles and Divisors on Riemann Surfaces

tant for Riemann surfaces (complex 1-manifolds). If 𝑀 is a compact Riemann

The association between line bundles and hypersurfaces is particularly impor-

surface, a closed hypersurface in 𝑀 is just a finite collection of points. A bundle

𝐿{𝑝} associated with a single point is called a point bundle.

as the tautological bundle on ℂℙ1 ) have no nontrivial holomorphic sections at all.

As we have seen, some holomorphic line bundles on Riemann surfaces (such

But we can get more information about the relationship between line bundles and

Let 𝑀 be a Riemann surface. A meromorphic function on 𝑀 is a holomorphic

hypersurfaces in this case by by expanding the type of sections we consider.

function 𝑓 ∶ 𝑀 ∖ 𝑃 → ℂ, where 𝑃 is a closed discrete subset of 𝑀, such that for

each 𝑝 ∈ 𝑃 , the function 1/𝑓 extends to a holomorphic function on a neighborhood
of 𝑝 that vanishes at 𝑝. Each 𝑝 ∈ 𝑃 is called a pole of 𝒇 , and the order of the zero
of 1/𝑓 at 𝑝 is called the order of the pole. To put it another way, for each 𝑝 ∈ 𝑀, if
𝑧 is a holomorphic coordinate centered at 𝑝, we can write 𝑓 in a punctured neigh-
borhood of 𝑝 in the form 𝑓 (𝑧) = 𝑧𝑘 ℎ(𝑧) for some integer 𝑘 and some nonvanishing
holomorphic function ℎ defined on a neighborhood of 𝑝; if 𝑘 > 0, then 𝑝 is a zero
of order 𝑘, and if 𝑘 < 0, then 𝑝 is a pole of order |𝑘|.
Similarly, if 𝐿 → 𝑀 is a holomorphic line bundle, a meromorphic section of
𝑳 is a holomorphic section defined on the complement of a closed discrete subset
𝑃 ⊆ 𝑀 such that for each 𝑝 ∈ 𝑃 , if 𝑠 is a holomorphic local frame for 𝐿 on a
neighborhood of 𝑝, then 𝜎(𝑧) = 𝑓 (𝑧)𝑠(𝑧) for some function 𝑓 that has a pole at 𝑝.
It is an easy exercise to check that in both cases, the order of a pole is independent

We will be primarily concerned with the case in which 𝑀 is compact. In that

of the choice of coordinates or local frame.

case, the set of poles of a meromorphic function or section is finite, as is the set of
zeros.
There is a simple algebraic construction that can be used to keep track of the

compact Riemann surface 𝑀, a divisor on 𝑀 is a finite formal linear combination

locations and orders of zeros and poles of meromorphic functions or sections. For a

of points of 𝑀 with integer coefficients (that is, an element of the free abelian group
on the set of points of 𝑀; see [LeeTM, p. 244]). A divisor 𝐷 = ∑𝑗 𝑛𝑗 𝑝𝑗 is said
to be effective if each of the integers 𝑛𝑗 is nonnegative. The set of divisors on 𝑀
forms an abelian group under addition, the free abelian group on the points of 𝑀.

If 𝑓 is a meromorphic function on 𝑀, we define the divisor of 𝒇 , denoted by

It is denoted by Div(𝑀).

(𝑓 ), to be the sum
𝑘 𝑙
(𝑓 ) = 𝑛𝑖 𝑝 𝑖 − 𝑚𝑗 𝑞𝑗 ,
∑ ∑
𝑖=1 𝑗=1
94 3. Holomorphic Vector Bundles

where 𝑝1 , … , 𝑝𝑘 are the zeros of 𝑓 , 𝑞1 , … , 𝑞𝑙 are its poles, and the integers 𝑛𝑖 and 𝑚𝑗
are their respective orders. A divisor of the form 𝐷 = (𝑓 ) for some meromorphic
function 𝑓 is called a principal divisor. The divisor of a product of meromorphic
functions is the sum of the divisors of the factors: (𝑓 𝑔) = (𝑓 ) + (𝑔), with the
understanding that a product like (𝑧 − 𝑎)𝑘 (𝑧 − 𝑎)𝑙 results in a removable singularity
at 𝑧 = 𝑎 if 𝑘 + 𝑙 ≥ 0. Therefore, the set of principal divisors forms a subgroup
of Div(𝑀). Two divisors are said to be linearly equivalent if their difference is a

is called the divisor class group of 𝑴, and denoted by Cl(𝑀).

principal divisor. The quotient of the group of divisors modulo principal divisors

The divisor of a meromorphic section 𝜎 of a holomorphic line bundle is de-

fined similarly and denoted by (𝜎). Such a divisor is effective if and only if 𝜎 is a
holomorphic section.
Here is a generalization of Theorem 3.39.

Theorem 3.41 (Line Bundle Associated with a Divisor). Suppose 𝑀 is a compact

Riemann surface. For each divisor 𝐷 ∈ Div(𝑀), there exist a holomorphic line
bundle 𝐿𝐷 → 𝑀, called the associated line bundle for 𝑫, and a meromorphic
section of 𝐿𝐷 , unique up to a constant multiple, whose divisor is equal to 𝐷. The

a meromorphic section with divisor equal to 𝐷 is isomorphic to 𝐿𝐷 . The map

bundle is unique in the sense that any other holomorphic line bundle that admits

sending 𝐷 to the isomorphism class of 𝐿𝐷 is a homomorphism from Div(𝑀) to

Pic(𝑀), whose kernel is the group of principal divisors; thus it descends to an
injective homomorphism from Cl(𝑀) to Pic(𝑀).

Proof. Given 𝐷 ∈ Div(𝑀), write 𝐷 = ∑𝑘𝛼=1 𝑛𝛼 𝑝𝛼 . For each 𝛼 such that 𝑛𝛼 ≠ 0,

choose a neighborhood 𝑈𝛼 of 𝑝𝛼 and a meromorphic function 𝑓𝛼 on 𝑈𝛼 that is
holomorphic and nonvanishing on 𝑈𝛼 ∖ {𝑝𝛼 } and has a zero of order 𝑛𝛼 if 𝑛𝛼 > 0
and a pole of order |𝑛𝛼 | if 𝑛𝛼 < 0. Let 𝑈0 = 𝑀 ∖ {𝑝1 , … , 𝑝𝑘 }, and 𝑓0 ≡ 1.
Whenever 𝑈𝛼 ∩ 𝑈𝛽 ≠ ∅, there is a nowhere-vanishing holomorphic function
𝜏𝛼𝛽 ∶ 𝑈𝛼 ∩ 𝑈𝛽 → GL(1, ℂ) such that 𝜏𝛼𝛽 = 𝑓𝛼 /𝑓𝛽 on the complement of {𝑝𝛼 , 𝑝𝛽 }.
As in the proof of Theorem 3.39, when 𝑈𝛼 ∩ 𝑈𝛽 ∩ 𝑈𝛾 ≠ ∅, we have 𝜏𝛼𝛽 𝜏𝛽𝛾 = 𝜏𝛼𝛾 .
The bundle construction theorem yields a holomorphic line bundle 𝐿𝐷 → 𝑀, with
a holomorphic local frame 𝑠𝛼 on each set 𝑈𝛼 and transition functions 𝜏𝛼𝛽 . The
meromorphic section 𝜎 is defined as before by setting 𝜎 = 𝑓𝛼 𝑠𝛼 on 𝑈𝛼 , and noting

that (𝜎) = 𝐷. If 𝜏 is another meromorphic section with the same divisor, then the
that these sections agree on overlaps. It follows immediately from the definition

ratio 𝜎/𝜏 has only removable singularities; thus it can be considered as a globally

If 𝐿′ → 𝑀 is another holomorphic line bundle that admits a meromorphic

defined holomorphic function, and is therefore constant.

shows that 𝐿′ ≅ 𝐿𝐷 .
section with the same divisor, the same argument as in the proof of Theorem 3.41
Applications of Holomorphic Line Bundles 95

To see that the map 𝐷 ↦ [𝐿𝐷 ] is a homomorphism, suppose 𝐷 and 𝐷′ are di-
visors. Then there are meromorphic sections 𝜎 of 𝐿𝐷 and 𝜎 ′ of 𝐿𝐷′ whose divisors
are 𝐷 and 𝐷′ , respectively. Then 𝜎 ⊗ 𝜎 ′ is a section of 𝐿𝐷 ⊗ 𝐿𝐷′ whose divisor
is 𝐷 + 𝐷′ , so the uniqueness argument above shows that 𝐿𝐷 ⊗ 𝐿𝐷′ ≅ 𝐿𝐷+𝐷′ .

𝐷 is in the kernel. That means 𝐿𝐷 is a trivial bundle, and it has a meromorphic

To verify the statement about the kernel of this homorphism, first suppose that

section 𝜎 whose divisor is 𝐷. Since a meromorphic section of the trivial bundle

is just a meromorphic function, this shows that 𝐷 is principal. Conversely, if 𝐷 is
principal, there is a global meromorphic function 𝑓 such that 𝐷 = (𝑓 ). Since a

the uniqueness statement in the first part of the proof shows that 𝐿𝐷 is isomorphic
meromorphic function is the same as a meromorphic section of the trivial bundle,

to the trivial bundle. □

We will show in Chapter 9 (see Thm. 9.61) that in fact the homomorphism from
Div(𝑀) to Pic(𝑀) is surjective, so it induces an isomorphism between the divisor
class group Cl(𝑀) and Pic(𝑀).
We should note that there is also a theory of divisors on higher-dimensional
complex manifolds; but because zeros and poles of meromorphic functions are no
longer isolated in that case, the theory is somewhat more complicated technically,
and we will not pursue it.

Line Bundles and Projective Embeddings

Our next major application of holomorphic line bundles will be as a tool for

Suppose 𝑀 is a compact complex manifold and 𝐿 → 𝑀 is a holomorphic

constructing embeddings into projective spaces.

line bundle. As Theorem 3.13 showed, the space 𝒪(𝑀; 𝐿) of global holomorphic
sections of 𝐿 is finite-dimensional. A point 𝑝 ∈ 𝑀 is called a base point for 𝑳
if every global holomorphic section of 𝐿 vanishes at 𝑝. The set of base points is
denoted by 𝐵(𝐿) and called the base locus of 𝑳.
If 𝐿 has at least one nontrivial holomorphic section, by choosing a basis
(𝑠0 , … , 𝑠𝑚 ) for 𝒪(𝑀; 𝐿) we can define a map from 𝑀 ∖ 𝐵(𝐿) to ℂℙ𝑚 as follows.
Given 𝑝0 ∈ 𝑀 ∖ 𝐵(𝐿), choose a local frame 𝑠 for 𝐿 in a neighborhood 𝑈 of 𝑝0 , ex-
press each section 𝑠𝑗 locally as 𝑠𝑗 = 𝑓𝑗 𝑠 for some holomorphic function 𝑓𝑗 ∈ 𝒪(𝑈 ),
and map each 𝑝 ∈ 𝑈 to the point [𝑓0 (𝑝), … , 𝑓𝑚 (𝑝)] ∈ ℂℙ𝑛 . To see that this does
not depend on the choice of local frame, let 𝑠 ̃ be any other local frame for 𝐿, so
there is a nonvanishing transition function 𝜏 satisfying 𝑠 ̃ = 𝜏𝑠 in a neighborhood
of 𝑝0 . If we write 𝑠𝑗 = 𝑓̃𝑗 𝑠,̃ then 𝑓̃𝑗 𝜏 = 𝑓𝑗 , and therefore [𝑓0 (𝑝), … , 𝑓𝑚 (𝑝)] =
[𝜏(𝑝)𝑓̃0 (𝑝), … , 𝜏(𝑝)𝑓̃𝑚 (𝑝)] = [𝑓̃0 (𝑝), … , 𝑓̃𝑚 (𝑝)]. Thus it makes sense to introduce
the notation [𝑠0 (𝑝), … , 𝑠𝑚 (𝑝)] to denote the point [𝑓0 (𝑝), … , 𝑓𝑚 (𝑝)] with respect
to any local frame, and define a holomorphic map 𝐹 ∶ 𝑀 ∖ 𝐵(𝐿) → ℂℙ𝑚 by
𝐹 (𝑝) = [𝑠0 (𝑝), … , 𝑠𝑚 (𝑝)]. Any two bases for 𝒪(𝑀; 𝐿) differ by a complex-linear
96 3. Holomorphic Vector Bundles

mation. Any one of these maps is called an associated map for 𝐿.

isomorphism, and therefore the corresponding maps differ by a projective transfor-

𝐿 → 𝑀 is a global embedding, based on the following lemma.

There are simple criteria for deciding when a map associated to a line bundle

Lemma 3.42. Suppose 𝑀 is a compact complex manifold and 𝐿 → 𝑀 is a holo-

Let 𝐹 ∶ 𝑀 ∖ 𝐵(𝐿) → ℂℙ𝑚 be an associated map.

morphic line bundle that admits at least one nontrivial global holomorphic section.

(a) Given distinct points 𝑝, 𝑞 ∈ 𝑀 ∖ 𝐵(𝐿), 𝐹 (𝑝) ≠ 𝐹 (𝑞) if and only if there
exists a holomorphic section 𝜎 of 𝐿 such that 𝜎(𝑝) = 0 and 𝜎(𝑞) ≠ 0.
(b) Given 𝑝 ∈ 𝑀 ∖ 𝐵(𝐿) and 𝑣 ∈ 𝑇𝑝′ 𝑀, 𝐷′ 𝐹 (𝑝)𝑣 ≠ 0 if and only if there is
a holomorphic section 𝜎 of 𝐿 such that 𝜎(𝑝) = 0 and 𝑣̂ 𝜎 (𝑝) ≠ 0, where
𝜎
̂ is the component function of 𝜎 with respect to some holomorphic local
frame.

Proof. If 𝑀 is a one-point space, then both claims are vacuously true, so assume
henceforth that 𝑀 contains at least two points. Choose a basis (𝑠0 , … , 𝑠𝑚 ) for
𝒪(𝑀; 𝐿) and write the associated map as 𝐹 (𝑝) = [𝑠0 (𝑝), … , 𝑠𝑚 (𝑝)].
To prove (a), let 𝑝, 𝑞 ∈ 𝑀 ∖ 𝐵(𝐿) be arbitrary distinct points, and suppose first
that there exists a section 𝜎 = ∑𝑚 𝑗=0 𝑎𝑗 𝑠𝑗 ∈ 𝒪(𝑀; 𝐿) with 𝜎(𝑝) = 0 and 𝜎(𝑞) ≠ 0.
It follows that 𝐹 (𝑝) lies in the projective hyperplane defined by the linear function
𝑓 (𝑤) = ∑𝑚 𝑗=0 𝑎𝑗 𝑤 but 𝐹 (𝑞) does not, so 𝐹 (𝑝) ≠ 𝐹 (𝑞).
𝑗

Conversely, suppose there is no section 𝜎 satisfying 𝜎(𝑝) = 0 and 𝜎(𝑞) ≠ 0.

Since 𝑝 is not a base point, there is some 𝑗 such that 𝑠𝑗 (𝑝) ≠ 0; after rearranging
the basis (which just changes 𝐹 by a projective transformation), we may assume
𝑠0 (𝑝) ≠ 0. For each 𝑘, let 𝜎𝑘 = 𝑠𝑘 − (𝑠𝑘 (𝑝)/𝑠0 (𝑝))𝑠0 , where the notation 𝑠𝑘 (𝑝)/𝑠0 (𝑝)
means the complex number 𝑏𝑘 such that 𝑠𝑘 (𝑝) = 𝑏𝑘 𝑠0 (𝑝). Then 𝜎𝑘 (𝑝) = 0 for each
𝑘, and our hypothesis implies that 𝜎𝑘 (𝑞) = 0 as well. This means

𝑠𝑘 (𝑝) 𝑠 (𝑞)
𝑠𝑘 (𝑞) = 𝑠0 (𝑞) = 𝜆𝑠𝑘 (𝑝) for each 𝑘, where 𝜆 = 0 .
( 𝑠0 (𝑝) ) 𝑠0 (𝑝)

Thus [𝑠0 (𝑞), … , 𝑠𝑚 (𝑞)] = [𝜆𝑠0 (𝑝), … , 𝜆𝑠𝑚 (𝑝)] = [𝑠0 (𝑝), … , 𝑠𝑚 (𝑝)], which means
𝐹 (𝑞) = 𝐹 (𝑝).
To prove (b), let 𝑝 ∈ 𝑀 ∖ 𝐵(𝐿) and 𝑣 ∈ 𝑇𝑝′ 𝑀 be arbitrary. As above, we can
arrange that 𝑠0 (𝑝) ≠ 0. We use 𝑠0 as a local frame for 𝐿 in a neighborhood of 𝑝, and
in that neighborhood write 𝑠𝑗 = 𝑓𝑗 𝑠0 for some holomorphic functions 𝑓0 , … , 𝑓𝑚
with 𝑓0 ≡ 1. Choose any holomorphic coordinates (𝑢𝑘 ) for 𝑀 on a neighborhood
of 𝑝, and use affine coordinates (𝑧1 , … , 𝑧𝑚 ) ↔ [1, 𝑧1 , … , 𝑧𝑚 ] on a neighborhood of
𝐹 (𝑝). Then 𝐹 has the coordinate representation 𝐹̂(𝑢) = (𝑓1 (𝑢), … , 𝑓𝑛 (𝑢)), and for
Applications of Holomorphic Line Bundles 97

any 𝑣 = ∑𝑘 𝑣𝑘 𝜕/𝜕𝑢𝑘 ∈ 𝑇𝑝′ 𝑀 we have

𝜕𝑓𝑗 𝜕 𝜕
𝐷′ 𝐹 (𝑝)𝑣 = (𝑝)𝑣𝑘 | = 𝑣(𝑓𝑗 ) 𝑗 | .
∑ 𝜕𝑢𝑘 𝜕𝑧 𝐹 (𝑝)
𝑗 ∑ 𝜕𝑧 𝐹 (𝑝)
(3.17)
𝑗,𝑘 𝑗

Suppose there is a section 𝜎 = ∑𝑚 𝑗=0 𝑎𝑗 𝑠𝑗 such that 𝜎(𝑝) = 0 and 𝑣(̂

𝜎 ) ≠ 0.
Note that our choice of 𝑠0 as a local frame implies
𝑚 𝑚
𝜎= 𝑎𝑠 = 𝑎𝑓 𝑠 ,
∑ 𝑗 𝑗 (∑ 𝑗 𝑗 ) 0
𝑗=0 𝑗=0

so the component function of 𝜎 is 𝜎 ̂ = ∑𝑚 𝑗=0 𝑎𝑗 𝑓𝑗 . The fact that 𝑣(̂

𝜎 ) ≠ 0 implies
that 𝑣(𝑓𝑗 ) ≠ 0 for some 𝑗 ≥ 1, and therefore (3.17) shows that 𝐷 𝐹 (𝑝)𝑣 ≠ 0.
′

Conversely, assume 𝐷′ 𝐹 (𝑝)𝑣 ≠ 0. Then (3.17) implies that

𝜕𝑓 𝑗
𝑣(𝑓𝑗 ) = 𝑣𝑘 (𝑝) ≠ 0 for some 𝑗,
𝜕𝑢𝑘
which proves that the condition of (b) is satisfied with 𝜎 = 𝑠𝑗 . □

Given a holomorphic line bundle 𝐿 over a complex manifold 𝑀, we say

𝒪(𝑀; 𝐿) separates points if for every pair of distinct points 𝑝, 𝑞 ∈ 𝑀, there exists
a global holomorphic section 𝜎 of 𝐿 such that 𝜎(𝑝) = 0 and 𝜎(𝑞) ≠ 0; and it sepa-
rates directions if for every 𝑝 ∈ 𝑀 and every nonzero 𝑣 ∈ 𝑇𝑝′ 𝑀, there is a global
holomorphic section 𝜎 such that 𝜎(𝑝) = 0 and 𝑣̂ 𝜎 (𝑝) ≠ 0, where 𝜎 ̂ is the compo-
nent function of 𝜎 with respect to some holomorphic local frame. (The reason for
insisting that 𝜎(𝑝) = 0 is because that guarantees the condition 𝑣̂ 𝜎 (𝑝) = 0 will be

A holomorphic line bundle 𝐿 over a compact complex manifold 𝑀 is said to

independent of the choice of local frame, as you can check.)

be very ample if
(i) 𝒪(𝑀; 𝐿) separates points, and
(ii) 𝒪(𝑀; 𝐿) separates directions.
(You will notice the strong resemblance to the definition of a Stein manifold. There
is no requirement analogous to holomorphic convexity in this case, though, because
it would be vacuous on a compact manifold. Compactness takes its place.)
Theorem 3.43. Suppose 𝑀 is a compact complex manifold and 𝐿 → 𝑀 is a
holomorphic line bundle. Each associated map for 𝐿 is a global embedding of 𝑀
into a projective space if and only 𝐿 is very ample.

Proof. First assume 𝑀 has dimension 0. Then 𝐿 is trivial, and we can write
𝑀 = {𝑝0 , … , 𝑝𝑘 }. For each 𝑘 there is a section 𝑠𝑘 that is nonzero at 𝑝𝑘 and zero
at the other points, and is vacuously holomorphic. Thus 𝐿 separates points, and it
vacuously separates directions because there are no nonzero tangent vectors. There-
fore, every line bundle is very ample, and every associated map is an embedding.
98 3. Holomorphic Vector Bundles

Assume henceforth that 𝑀 has positive dimension. Suppose 𝐿 → 𝑀 is a

very ample holomorphic line bundle. First we note that 𝐿 has no base points. Let
𝑞 ∈ 𝑀 be arbitrary. Then for any other point 𝑝 ≠ 𝑞, there is a holomorphic section
𝜎 ∶ 𝑀 → 𝐿 such that 𝜎(𝑝) = 0 and 𝜎(𝑞) ≠ 0, so not all sections vanish at 𝑞. Thus
the associated map 𝐹 is defined on all of 𝑀. It follows from the definition of very
ample and Lemma 3.42 that 𝐹 is an injective holomorphic immersion into ℂℙ𝑛 ,
and the closed map lemma [LeeTM, Lemma 4.50] shows that 𝐹 is an embedding.

𝐹 ∶ 𝑀 → ℂℙ𝑛 . The fact that 𝐹 is globally defined means that 𝐿 has no base
Conversely, suppose the associated map is a global holomorphic embedding

points. Because 𝐹 is an injective holomorphic immersion, Lemma 3.42 shows that

𝒪(𝑀; 𝐿) separates points and directions, so 𝐿 is very ample. □

Somewhat more generally, a holomorphic line bundle 𝐿 is said to be ample

if for some positive integer 𝑘, the tensor power 𝐿𝑘 is very ample. The reason for
this seemingly trivial distinction is that, as we will see in Chapter 10, there is a
relatively simple criterion for determining whether a particular bundle is ample,
whereas determining if it is very ample is not so straightforward.
Corollary 3.44. A compact complex manifold is projective if and only if it admits
an ample holomorphic line bundle.

Proof. A compact 0-dimensional manifold certainly admits an embedding into pro-

jective space, and the only holomorphic line bundle on it is the trivial one, which is

that 𝑀 is a compact complex manifold of positive dimension.

very ample by the argument in the proof of Theorem 3.43. So assume henceforth

If 𝑀 admits an ample line bundle 𝐿, then Theorem 3.43 shows that the asso-
ciated map for some positive power of 𝐿 is a holomorphic embedding.
Conversely, suppose 𝑀 is projective, so it is biholomorphic to a complex sub-
manifold of ℂℙ𝑛 for some 𝑛; we may as well assume 𝑀 is itself a compact complex
submanifold of ℂℙ𝑛 . Let [𝑤0 , … , 𝑤𝑛 ] denote homogeneous coordinates on ℂℙ𝑛 ,
and let 𝐿 → 𝑀 be the restriction to 𝑀 of the hyperplane bundle 𝐻 → ℂℙ𝑛 . We
will show that 𝐿 is very ample.
To show that 𝒪(𝑀; 𝐿) separates points, suppose 𝑝, 𝑞 are distinct points in 𝑀,
represented in homogeneous coordinates by [𝑝0 , … , 𝑝𝑛 ] and [𝑞 0 , … , 𝑞 𝑛 ], respec-
tively. Because the vectors 𝑃 = (𝑝0 , … , 𝑝𝑛 ) and 𝑄 = (𝑞 0 , … , 𝑞 𝑛 ) are linearly
independent in ℂ𝑛+1 , there is a linear function 𝑓 ∶ ℂ𝑛+1 → ℂ such that 𝑓 (𝑃 ) = 0
and 𝑓 (𝑄) ≠ 0. Let 𝜎 = 𝜑𝑓 |𝑀 (using the notation of (3.15)); this is a holomorphic
section of 𝐿 that satisfies 𝜎(𝑝) = 0 and 𝜎(𝑞) ≠ 0.
To show that 𝒪(𝑀; 𝐿) separates directions, let 𝑝 ∈ 𝑀 and 𝑣 ∈ 𝑇𝑝′ 𝑀 with
𝑣 ≠ 0. There is some affine coordinate chart containing 𝑝; after renumbering the
coordinates if necessary, we may assume it is the chart 𝑈0 defined by 𝑤0 ≠ 0,
and write 𝑝 = [1, 𝑝1 , … , 𝑝𝑛 ]. In affine coordinates (𝑧1 , … , 𝑧𝑛 ) ↔ [1, 𝑧1 , … , 𝑧𝑛 ],
Problems 99

we can write 𝑣 = 𝑣𝑗 𝜕/𝜕𝑧𝑗 |𝑝 , and since 𝑣 ≠ 0, there is some component 𝑣𝑘 that is

nonzero. Let 𝑓 ∶ ℂ𝑛+1 → ℂ be the linear function 𝑓 (𝑤0 , … , 𝑤𝑛 ) = 𝑤𝑘 − 𝑝𝑘 𝑤0 , let
𝜑𝑓 ∶ ℂℙ𝑛 → 𝐻 be the corresponding section, and let 𝜎 = 𝜑𝑓 |𝑀 . Then 𝜎(𝑝) = 0.
The section 𝜑𝑤0 ∶ ℂℙ𝑛 → 𝐻 is nonvanishing on 𝑈0 , so it gives a local frame for
𝐻 there, and it restricts to a local frame for 𝐿. The component function of 𝜎 with
respect to this frame is
𝑤𝑘 − 𝑝 𝑘 𝑤0
𝜎
̂([𝑤]) = ,
𝑤0
which has the coordinate representation

𝜎
̂(𝑧1 , … , 𝑧𝑛 ) = 𝑧𝑘 − 𝑝𝑘 ,

so 𝑣̂
𝜎 = 𝑣𝑘 ≠ 0. □

The next proposition describes an important property of very ample line bun-
dles.
Proposition 3.45. Suppose 𝑀 is a compact complex manifold, 𝐿 → 𝑀 is a very
ample holomorphic line bundle, and 𝐹 ∶ 𝑀 → ℂℙ𝑛 is its associated map. Then
𝐿 ≅ 𝐹 ∗ 𝐻, where 𝐻 → ℂℙ𝑛 is the hyperplane bundle.

Proof. Problem 3-13. □

Problems
3-1. Let 𝐻 → ℂℙ𝑛 be the hyperplane bundle. For any distinct integers 𝑘, 𝑙,
show that the tensor powers 𝐻 𝑘 and 𝐻 𝑙 are not isomorphic to each other.
3-2. Let 𝑚 > 𝑛 and let 𝐹 ∶ ℂℙ𝑛 → ℂℙ𝑚 be the holomorphic embedding

𝐹 ([𝑤0 , … , 𝑤𝑛 ]) = [𝑤0 , … , 𝑤𝑛 , 0, … , 0].

Let 𝐻 → ℂℙ𝑚 denote the hyperplane bundle of ℂℙ𝑚 . Show that 𝐹 ∗ 𝐻 is

isomorphic to the hyperplane bundle of ℂℙ𝑛 .
3-3. Let 𝐻 → ℂℙ1 be the hyperplane bundle. Show that 𝐻 2 ≅ 𝑇 ′ ℂℙ1 .
3-4. Let 𝑀 be a complex manifold. A holomorphic vector field on 𝑀 is a
holomorphic section of 𝑇 ′ 𝑀. Let 𝑍 be a smooth section of 𝑇 ′ 𝑀 and let
𝜃𝑡 denote the flow of Re 𝑍. Show that 𝑍 is holomorphic if and only if 𝜃𝑡
is a holomorphic map (where it is defined) for each 𝑡.
3-5. For 𝑛 ≥ 1, let 𝑈0 ⊆ ℂℙ𝑛 be the image of the standard affine embedding
(𝑧1 , … , 𝑧𝑛 ) ↦ [1, 𝑧1 , … , 𝑧𝑛 ]. Show that there are global holomorphic
vector fields 𝑍1 , … , 𝑍𝑛 on ℂℙ𝑛 whose coordinate representations in 𝑈0
are 𝑍𝑗 = 𝜕/𝜕𝑧𝑗 .
100 3. Holomorphic Vector Bundles

3-6. Suppose 𝑀 is a complex manifold. Let 𝜋 ∶ 𝑀 ̃ → 𝑀 be the blowup

of 𝑀 at a point 𝑝 ∈ 𝑀, and let 𝐸 = 𝜋 (𝑝) ⊆ 𝑀
−1 ̃ be the exceptional

(a) Show that if 𝑁 is another complex manifold and 𝐹 ∶ 𝑀 ̃ → 𝑁 is a

hypersurface.

holomorphic map that is constant on 𝐸, then there is a unique holo-

morphic map 𝑓 ∶ 𝑀 → 𝑁 such that 𝑓 ∘ 𝜋 = 𝐹 .
(b) Show that the pullback 𝜋 ∗ ∶ 𝒪(𝑀) → 𝒪(𝑀) ̃ is an isomorphism.

3-7. Suppose 𝑀 and 𝑁 are complex manifolds and 𝐹 ∶ 𝑀 → 𝑁 is a holomor-

phic map. Show that the global holomorphic Jacobian map 𝐷′ ∶ 𝑇 ′ 𝑀 →
𝑇 ′ 𝑁 is holomorphic.
3-8. Suppose 𝐺 is a connected, compact, complex Lie group. Prove that 𝐺 is
abelian, as follows. Let 𝔤 be the Lie algebra of left-invariant vector fields
on 𝐺, and let 𝑒 ∈ 𝐺 be the identity.
(a) Let 𝐶 ∶ 𝐺 × 𝐺 → 𝐺 be the holomorphic map 𝐶(𝑔, ℎ) = 𝑔ℎ𝑔 −1 . For
each 𝑔 ∈ 𝐺, the map 𝐶𝑔 ∶ 𝐺 → 𝐺 defined by 𝐶𝑔 (ℎ) = 𝐶(𝑔, ℎ) is
called conjugation by 𝒈. Show that 𝐷(𝐶𝑔 )(𝑒)𝑣 = 𝐷𝐶(𝑔, 𝑒)(0, 𝑣) for
each 𝑣 ∈ 𝑇 ′ 𝐺.
(b) Show that the map from 𝐺 to GL(𝑇𝑒′ 𝐺) given by 𝑔 ↦ 𝐷′ (𝐶𝑔 )(𝑒) is
holomorphic, and conclude that 𝐷′ (𝐶𝑔 )(𝑒) = Id for all 𝑔.
(c) The adjoint representation of 𝑮 is the map Ad ∶ 𝐺 → GL(𝔤) given
by Ad(𝑔) = (𝐶𝑔 )∗ (the Lie algebra isomorphism induced by the group
isomorphism 𝐶𝑔 ). Show that Ad(𝑔) = Id for all 𝑔 ∈ 𝐺.
(d) The adjoint representation of 𝖌 is the map ad ∶ 𝔤 → End(𝔤) given
by ad(𝑋)(𝑌 ) = [𝑋, 𝑌 ]. Using the fact that Ad∗ = ad [LeeSM, Thm.
20.27], show that 𝔤 is abelian and therefore 𝐺 is abelian.
3-9. Let 𝑇 → ℂℙ𝑛 be the tautological bundle. Show that the total space 𝑇 is
diffeomorphic to ℂℙ𝑛+1 minus a point, by considering the smooth map
𝐹 ∶ 𝑇 → ℂℙ𝑛+1 given by

𝐹 ([𝑤], 𝑣) = [𝑤0 , … , 𝑤𝑛 , 𝑤 ⋅ 𝑣],

where 𝑤 ⋅ 𝑣 = ∑𝑛𝑗=0 𝑤𝑗 𝑣𝑗 .
3-10. If 𝑀 and 𝑁 are connected smooth oriented 𝑛-manifolds, their oriented
connected sum is a manifold 𝑀 # 𝑁 obtained as follows. Choose a posi-
tively oriented smooth coordinate chart (𝑈 , 𝜑) for 𝑀 and a negatively ori-
ented one (𝑉 , 𝜓) for 𝑁, and choose 𝜀 > 0 such that both 𝜑(𝑈 ) and 𝜓(𝑉 )
contain 𝐵2𝜀 (0). Let 𝑀 ′ = 𝑀 ∖𝜑−1 (𝐵 𝜀/2 (0)), 𝑁 ′ = 𝑁 ∖𝜓 −1 (𝐵 𝜀/2 (0)), and
let 𝑀 #𝑁 be the quotient space obtained from the disjoint union 𝑀 ′ ⨿𝑁 ′
by identifying 𝜑−1 (𝑥) with 𝜓 −1 (𝜀2 𝑥/|𝑥|2 ) for all 𝑥 ∈ 𝐵2𝜀 (0) ∖ 𝐵 𝜀/2 (0).
Then 𝑀 # 𝑁 is a connected 𝑛-manifold with a unique smooth structure
Problems 101

embeddings of 𝑀 ′ and 𝑁 ′ into 𝑀 #𝑁, and an orientation consistent with

determined by the requirement that the quotient map restricts to smooth

those of 𝑀 ′ and 𝑁 ′ (you do not need to prove this).

Suppose 𝑀 is a connected complex 𝑛-manifold and 𝑀 ̃ is the blowup of
𝑀 at a point. Use the result of Problem 3-9 to show that 𝑀 ̃ is diffeo-
morphic to the oriented connected sum 𝑀 # ℂℙ , where ℂℙ denotes the
𝑛 𝑛

smooth manifold ℂℙ𝑛 with the opposite orientation.

3-11. Let 𝑀 be a complex manifold. Suppose 𝑆1 , … , 𝑆𝑚 ⊆ 𝑀 are disjoint
closed complex hypersurfaces and 𝑆 = 𝑆1 ∪ ⋯ ∪ 𝑆𝑚 . Show that the line
bundle 𝐿𝑆 associated with 𝑆 is isomorphic to 𝐿𝑆1 ⊗ ⋯ ⊗ 𝐿𝑆𝑚 .
3-12. Suppose 𝑀 ̃ and 𝑀 are complex manifolds, 𝑆 ⊆ 𝑀 is a closed complex
hypersurface, and 𝑓 ∶ 𝑀 ̃ → 𝑀 is a holomorphic map that is transverse
to 𝑆, so that 𝑆̃ = 𝑓 (𝑆) is a closed complex hypersurface in 𝑀
−1 ̃ (see
Example 2.23). Show that 𝑓 𝐿𝑆 ≅ 𝐿𝑆̃. [Hint: Consider a holomorphic
∗

section of 𝐿𝑆 that vanishes simply on 𝑆.]

3-13. Prove Proposition 3.45 (a very ample line bundle is isomorphic to the

3-14. Show that the action of ℤ/2 on ℂ𝑛 generated by 𝑧 ↦ −𝑧 lifts to a free,

pullback of the hyperplane bundle).

proper, and holomorphic action on the blowup of ℂ𝑛 at the origin.

 10.1090/gsm/244/04

Chapter 4

The Dolbeault Complex

For a complex manifold 𝑀, the decomposition of 𝑇ℂ 𝑀 into 𝑇 ′ 𝑀 ⊕ 𝑇 ″ 𝑀 leads

to a related decomposition of all complex-valued differential forms, from which we
can build new biholomorphic invariants called the Dolbeault cohomology groups.

Decomposing Differential Forms by Type

Suppose 𝑀 is a complex 𝑛-manifold. For 0 ≤ 𝑘 ≤ 𝑛, we define the bundle of
complex 𝒌-forms, denoted by Λ𝑘ℂ 𝑀, to be the complexification of Λ𝑘 𝑀. We can
view the fiber of Λ𝑘ℂ 𝑀 at a point 𝑎 ∈ 𝑀 as the space of alternating complex-
multilinear maps from (𝑇𝑎 𝑀)ℂ to ℂ. Just as for any complexified bundle, every
smooth section of Λ𝑘ℂ 𝑀 can be written uniquely as a sum 𝜔 + 𝑖𝜂, where 𝜔 and 𝜂 are
ordinary smooth real 𝑘-forms, and Λ𝑘ℂ 𝑀 has a natural conjugation operator given
by 𝜔 + 𝑖𝜂 = 𝜔 − 𝑖𝜂. A complex differential form 𝜔 is said to be real if 𝜔 = 𝜔. The
exterior derivative operator 𝑑 extends immediately by complex linearity to complex
differential forms, and it still satisfies the antiderivation rule 𝑑(𝛼 ∧ 𝛽) = 𝑑𝛼 ∧ 𝛽 +
(−1)𝑘 𝛼 ∧ 𝑑𝛽 when 𝛼 is a complex 𝑘-form and 𝛽 is a complex 𝑙-form. The integral
of a complex 𝑘-form over a 𝑘-dimensional (real) manifold is defined by integrating

of the 𝑘-form is compact, and Stokes’s theorem holds for such forms.
the real and imaginary parts separately, provided either the manifold or the support

The complex structure on 𝑇 𝑀 yields another way to decompose complex

any holomorphic local coordinates (𝑧1 , … , 𝑧𝑛 ), the 1-forms (𝑑𝑧1 , … , 𝑑𝑧𝑛 , 𝑑𝑧1 , … ,
forms, which is more useful than real and imaginary parts. In the domain of

𝑑𝑧𝑛 ) provide a local frame for the complexified cotangent bundle. Thus the follow-
ing collection of forms constitutes a smooth local frame for Λ𝑘ℂ 𝑀:

{𝑑𝑧 1 ∧ ⋯ ∧ 𝑑𝑧 𝑝 ∧ 𝑑𝑧 1 ∧ ⋯ ∧ 𝑑𝑧 𝑞 ∶
𝑗 𝑗 𝑙 𝑙

𝑝 + 𝑞 = 𝑘, 𝑗1 < ⋯ < 𝑗𝑝 , and 𝑙1 < ⋯ < 𝑙𝑞 }.

103
104 4. The Dolbeault Complex

each term has exactly 𝑝 of the 𝑑𝑧𝑗 factors and 𝑞 of the 𝑑𝑧𝑙 factors when it is ex-
We wish to separate out the complex differential forms with the property that

pressed in terms of holomorphic local coordinates. But to make this definition

meaningful, we need to verify that it is independent of coordinates. That follows
from the next lemma.
Lemma 4.1. Suppose 𝛼 is a complex 𝑘-form on a complex manifold, and 𝑝 + 𝑞 = 𝑘.
Then the following are equivalent:
(a) In every local holomorphic coordinate chart (𝑧1 , … , 𝑧𝑛 ), 𝛼 can be ex-
pressed as a sum of terms, each of which has exactly 𝑝 of the 𝑑𝑧𝑗 factors
and 𝑞 of the 𝑑𝑧𝑙 factors.
(b) If 𝑉1 , … , 𝑉𝑘 are complex vector fields on an open subset of 𝑀, then
𝛼(𝑉1 , … , 𝑉𝑘 ) = 0 if more than 𝑝 of the vector fields are sections of 𝑇 ′ 𝑀,
or more than 𝑞 of them are sections of 𝑇 ″ 𝑀.

► Exercise 4.2. Prove this lemma.

Thus for 𝑝 + 𝑞 = 𝑘, we say a complex 𝑘-form is of type (𝒑, 𝒒) or bidegree

(𝒑, 𝒒), or is a (𝒑, 𝒒)-form for short, if it satisfies either of the equivalent conditions
in the preceding lemma. Let Λ𝑝,𝑞 𝑀 ⊆ Λ𝑘ℂ 𝑀 be the subset consisting of forms of
type (𝑝, 𝑞). Lemma 4.1 shows that Λ𝑝,𝑞 𝑀 is locally spanned by smooth sections of
Λ𝑘ℂ 𝑀, so it is a smooth subbundle. Moreover, because every complex 𝑘-form is a
sum of forms of type (𝑝, 𝑞) for 𝑝 + 𝑞 = 𝑘, and Λ𝑝,𝑞 𝑀 ∩ Λ𝑝 ,𝑞 𝑀 contains only the
′ ′

zero form unless 𝑝 = 𝑝′ and 𝑞 = 𝑞 ′ , we have a Whitney sum decomposition

Λ𝑘ℂ 𝑀 = Λ𝑝,𝑞 𝑀.
⨁
𝑝+𝑞=𝑘

Thus for each (𝑝, 𝑞) there is a coordinate-independent projection operator

𝜋 𝑝,𝑞 ∶ Λ𝑘ℂ 𝑀 → Λ𝑝,𝑞 𝑀.
We use the notation ℰ 𝑘 (𝑀) to denote the space of smooth sections of Λ𝑘ℂ 𝑀, and
ℰ 𝑝,𝑞 (𝑀) for the space of smooth sections of Λ𝑝,𝑞 𝑀. By convention, ℰ 𝑝,𝑞 (𝑀) = 0
on a complex 𝑛-manifold except for 0 ≤ 𝑝, 𝑞 ≤ 𝑛, and similarly ℰ 𝑘 (𝑀) = 0 except
for 0 ≤ 𝑘 ≤ 2𝑛.
Proposition 4.3. Let 𝑀 be a complex manifold, 𝛼 ∈ ℰ 𝑝,𝑞 (𝑀), and 𝛽 ∈ ℰ 𝑝 ,𝑞 ′
(𝑀).
′

(a) 𝛼 ∈ ℰ 𝑞,𝑝 (𝑀).

(b) 𝛼 ∧ 𝛽 ∈ ℰ 𝑝+𝑝 ,𝑞+𝑞 ′
(𝑀).
′

► Exercise 4.4. Prove this proposition.

The real usefulness of the decomposition of forms by types is based on the

following proposition.
Decomposing Differential Forms by Type 105

Proposition 4.5. If 𝑀 is a complex manifold, then

𝑑(ℰ 𝑝,𝑞 (𝑀)) ⊆ ℰ 𝑝+1,𝑞 (𝑀) ⊕ ℰ 𝑝,𝑞+1 (𝑀).

Proof. Suppose 𝛼 ∈ ℰ 𝑝,𝑞 (𝑀). This is a local question, so we may choose holo-
morphic local coordinates (𝑧1 , … , 𝑧𝑛 ) and write
′
𝛼= 𝛼𝐽 𝐿 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑝 ∧ 𝑑𝑧𝑙1 ∧ ⋯ ∧ 𝑑𝑧𝑙𝑞 ,
∑
𝐽 ,𝐿

where 𝐽 = (𝑗1 , … , 𝑗𝑝 ) and 𝐿 = (𝑙1 , … , 𝑙𝑞 ) are multi-indices, and the primed sum-

satisfy 𝑗1 < ⋯ < 𝑗𝑝 and 𝑙1 < ⋯ < 𝑙𝑞 . It follows that

mation sign denotes a sum over only increasing multi-indices, that is, those that

′ 𝜕𝛼𝐽 𝐿 𝑟 𝜕𝛼𝐽 𝐿 𝑟
𝑑𝛼 = 𝑑𝑧 + 𝑑𝑧 ) ∧ 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑝 ∧ 𝑑𝑧𝑙1 ∧ ⋯ ∧ 𝑑𝑧𝑙𝑞 ,
∑ ∑( 𝜕𝑧𝑟 𝜕𝑧 𝑟
𝐽 ,𝐿 𝑟

which expands to a sum of (𝑝 + 1, 𝑞)-forms and (𝑝, 𝑞 + 1)-forms. □

to let 𝑀 be a complex 𝑛-manifold. For each 𝑝, 𝑞 ∈ {0, … , 𝑛}, define the Dolbeault
Thanks to this proposition, we can make the following definitions. We continue

operator 𝜕 ∶ ℰ 𝑝,𝑞 (𝑀) → ℰ 𝑝,𝑞+1 (𝑀) and its conjugate 𝜕 ∶ ℰ 𝑝,𝑞 (𝑀) → ℰ 𝑝+1,𝑞 (𝑀)

𝜕 = 𝜋 𝑝,𝑞+1 ∘ 𝑑, 𝜕 = 𝜋 𝑝+1,𝑞 ∘ 𝑑.
by

More generally, if 𝛼 is any complex differential form, we define 𝜕𝛼 and 𝜕𝛼 by de-

composing 𝛼 into terms of type (𝑝, 𝑞) and applying the formulas above to each term
separately. These operators, first introduced by the English mathematician William
V. D. Hodge [Hod41], were later named after Pierre Dolbeault in recognition of his
use of them in proving the theorem now known as the Dolbeault theorem (Thm. 6.19
below).
Example 4.6 (Dolbeault Operators in Coordinates). If 𝑢 is a smooth complex-
valued function (a (0, 0)-form), we have the following formulas in holomorphic
coordinates (using the summation convention):
𝜕𝑢 𝑗 𝜕𝑢 𝑗
𝜕𝑢 = 𝑑𝑧 , 𝜕𝑢 = 𝑑𝑧 .
𝜕𝑧𝑗 𝜕𝑧𝑗
(4.1)

And for a (1, 0)-form 𝛼 = 𝛼𝑗 𝑑𝑧𝑗 ,

𝜕𝛼𝑗 𝜕𝛼𝑗
𝜕𝛼 = 𝑑𝑧𝑙 ∧ 𝑑𝑧𝑗 , 𝜕𝛼 = 𝑑𝑧𝑙 ∧ 𝑑𝑧𝑗 .
𝜕𝑧𝑙 𝜕𝑧𝑙
More generally, if
′
𝛼= 𝛼𝐽 𝐿 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑝 ∧ 𝑑𝑧𝑙1 ∧ ⋯ ∧ 𝑑𝑧𝑙𝑞 ,
∑
𝐽 ,𝐿
106 4. The Dolbeault Complex

𝜕𝛼𝐽 𝐿 𝑟
then
′
𝜕𝛼 = 𝑑𝑧 ∧ 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑝 ∧ 𝑑𝑧𝑙1 ∧ ⋯ ∧ 𝑑𝑧𝑙𝑞 , and
∑ ∑ 𝜕𝑧𝑟
(4.2)
𝐽 ,𝐿 𝑟
′ 𝜕𝛼𝐽 𝐿 𝑟
𝜕𝛼 = 𝑑𝑧 ∧ 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑝 ∧ 𝑑𝑧𝑙1 ∧ ⋯ ∧ 𝑑𝑧𝑙𝑞 ,
∑ ∑ 𝜕𝑧𝑟
(4.3)
𝐽 ,𝐿 𝑟

as you can check. //

The operator 𝜕 is also sometimes called the Cauchy–Riemann operator, in part
because of the following result.
Proposition 4.7. On a complex manifold, a smooth function 𝑓 is holomorphic if
and only if 𝜕𝑓 ≡ 0.

Proof. This follows immediately from (4.1) and Proposition 1.42(a). □

Proposition 4.8. On a complex manifold 𝑀, if 𝛼 ∈ ℰ 𝑘 (𝑀) and 𝛽 ∈ ℰ 𝑙 (𝑀), then

𝜕(𝛼 ∧ 𝛽) = 𝜕𝛼 ∧ 𝛽 + (−1)𝑘 𝛼 ∧ 𝜕𝛽, 𝜕(𝛼 ∧ 𝛽) = 𝜕𝛼 ∧ 𝛽 + (−1)𝑘 𝛼 ∧ 𝜕𝛽.

Proof. This follows from Propositions 4.5 and 4.3. □

Proposition 4.9. Let 𝑀 be a complex manifold. For every complex differential

form 𝛼 on 𝑀, the following identities hold:
(4.4) 𝑑𝛼 = 𝜕𝛼 + 𝜕𝛼,
𝜕𝛼 = 𝜕(𝛼),
𝜕𝜕𝛼 = 𝜕𝜕𝛼 = 0,
(4.5)

𝜕𝜕𝛼 = −𝜕𝜕𝛼.
(4.6)
(4.7)

Proof. By decomposing 𝛼 into types and working with each type separately, we see
that it suffices to prove these identities under the assumption that 𝛼 is a (𝑝, 𝑞)-form.
Equation (4.4) follows directly from Proposition 4.5 and the definition of 𝜕 and 𝜕,
and (4.5) follows from the coordinate formulas (4.2) and (4.3) because 𝑑𝑧𝑗 and 𝑑𝑧𝑗
are conjugates of each other.
For (4.6) and (4.7), note that
0 = 𝑑(𝑑𝛼) = (𝜕 + 𝜕)(𝜕 + 𝜕)𝛼 = 𝜕𝜕𝛼 + (𝜕𝜕𝛼 + 𝜕𝜕𝛼) + 𝜕𝜕𝛼.
On the right-hand side, the first term is in ℰ 𝑝+2,𝑞 (𝑀), the term in parentheses is in
ℰ 𝑝+1,𝑞+1 (𝑀), and the last term is in ℰ 𝑝,𝑞+2 (𝑀). Since these spaces intersect only
in the zero form, each of those three terms must be zero. □

The importance of the Dolbeault operators stems from the fact that they are
preserved by holomorphic maps.
Decomposing Differential Forms by Type 107

Proposition 4.10. Suppose 𝑀 and 𝑁 are complex manifolds and 𝐹 ∶ 𝑀 → 𝑁 is

a holomorphic map. Then for all 𝑝 and 𝑞 and all 𝛼 ∈ ℰ 𝑝,𝑞 (𝑁),
(4.8) 𝐹 ∗ (ℰ 𝑝,𝑞 (𝑁)) ⊆ ℰ 𝑝,𝑞 (𝑀),
(4.9) 𝐹 ∗ (𝜕𝛼) = 𝜕(𝐹 ∗ 𝛼),
(4.10) 𝐹 ∗ (𝜕𝛼) = 𝜕(𝐹 ∗ 𝛼).

Proof. Again, these are local assertions, so for each 𝑝 ∈ 𝑀 we may choose holo-
morphic coordinates (𝑧1 , … , 𝑧𝑚 ) on a neighborhood of 𝑝 and (𝑤1 , … , 𝑤𝑛 ) on a
neighborhood of 𝐹 (𝑝), and compute
𝜕𝐹 𝑗 𝑙 𝜕𝐹 𝑗 𝑙
𝐹 ∗ 𝑑𝑤𝑗 = 𝑑𝑧 , 𝐹 ∗
𝑑𝑤 𝑗
= 𝑑𝑧 .
𝜕𝑧𝑙 𝜕𝑧𝑙
Inserting these into the coordinate formula for 𝐹 ∗ 𝛼 when 𝛼 is a (𝑝, 𝑞)-form shows
that the number of 𝑑𝑧𝑗 and 𝑑𝑧𝑗 factors in each term of 𝐹 ∗ 𝛼 is exactly the same as the
number of 𝑑𝑤𝑗 and 𝑑𝑤𝑗 factors, respectively, in the expression for 𝛼, so 𝐹 ∗ 𝛼 is also
a (𝑝, 𝑞)-form. This proves (4.8). It follows from this that 𝐹 ∗ ∘ 𝜋 𝑝,𝑞 = 𝜋 𝑝,𝑞 ∘ 𝐹 ∗ , and
then (4.9) and (4.10) follow because 𝐹 ∗ commutes with 𝑑 and with the projections
𝜋 𝑝+1,𝑞 and 𝜋 𝑝,𝑞+1 . □

This allows us to define a new set of biholomorphic invariants. First, here are a
few definitions that may be familiar from algebraic topology. A cochain complex

integers, {𝐴𝑞 ∶ 𝑞 ∈ ℤ}, together with homomorphisms 𝑑 𝑞 ∶ 𝐴𝑞 → 𝐴𝑞+1 for each

is a sequence of abelian groups (or real or complex vector spaces) indexed by the

𝑞, such that the composition of any two successive homomorphisms is zero. (Fre-
quently in practice, the groups are defined only for values of 𝑞 in a certain range,
in which case we just take 𝐴𝑞 = 0 for other values of 𝑞.) We often denote such a
complex by 𝐴∗ , with the homomorphisms understood from the context. The 𝒒th
cohomology group of 𝐴∗ , denoted by 𝐻 𝑞 (𝐴∗ ), is the quotient of the kernel of the
𝑞th homomorphism by the image of the previous one:
Ker (𝑑 𝑞 ∶ 𝐴𝑞 → 𝐴𝑞+1 )
𝐻 𝑞 (𝐴∗ ) = .
Im (𝑑 𝑞−1 ∶ 𝐴𝑞−1 → 𝐴𝑞 )
If 𝐴∗ is a cochain complex of vector spaces, then the homology groups are objects in
the same category. If 𝐴∗ and 𝐵 ∗ are cochain complexes, a cochain map 𝜑 ∶ 𝐴∗ →
𝐵 ∗ is a collection of homomorphisms 𝜑𝑞 ∶ 𝐴𝑞 → 𝐵 𝑞 that satisfy 𝜑𝑞+1 ∘ 𝑑 𝑞 =
𝑑 𝑞 ∘ 𝜑𝑞 for all 𝑞; any such map descends to a homomorphism 𝜑∗ ∶ 𝐻 𝑞 (𝐴∗ ) →
𝐻 𝑞 (𝐵 ∗ ), called the induced cohomology homomorphism. (For completeness, we

vector spaces) 𝐴∗ = {𝐴𝑞 ∶ 𝑞 ∈ ℤ}, with homomorphisms 𝜕𝑞 ∶ 𝐴𝑞 → 𝐴𝑞−1

remark that a chain complex is a sequence of abelian groups (or real or complex

going in the direction of decreasing indices and satisfying 𝜕𝑞−1 ∘ 𝜕𝑞 = 0. The

chain complex, denoted by 𝐻𝑞 (𝐴∗ ). A chain map between chain complexes 𝐴∗ and
corresponding quotient groups in that case are called the homology groups of the
108 4. The Dolbeault Complex

𝐵∗ is a collection of homomorphisms 𝜑𝑞 ∶ 𝐴𝑞 → 𝐵𝑞 satisfying 𝜑𝑞−1 ∘ 𝜕𝑞 = 𝜕𝑞 ∘ 𝜑𝑞 ,

which descend to homomorphisms 𝜑∗ ∶ 𝐻𝑞 (𝐴∗ ) → 𝐻𝑞 (𝐵∗ ).)
Now let 𝑀 be a complex 𝑛-manifold. Because 𝜕 ∘ 𝜕 ≡ 0, for each 𝑝 we obtain
a cochain complex called the 𝒑th Dolbeault complex:
𝜕 𝜕 𝜕
0 → ℰ 𝑝,0 (𝑀) −
→ ℰ 𝑝,1 (𝑀) −
→⋯−
→ ℰ 𝑝,𝑛 (𝑀) → 0.

Then we define the Dolbeault cohomology groups 𝐻 𝑝,𝑞 (𝑀) as the cohomology
groups of this complex, which are the following complex vector spaces:

Ker (𝜕 ∶ ℰ 𝑝,𝑞 (𝑀) → ℰ 𝑝,𝑞+1 (𝑀))

𝐻 𝑝,𝑞
(𝑀) = .
Im (𝜕 ∶ ℰ 𝑝,𝑞−1 (𝑀) → ℰ 𝑝,𝑞 (𝑀))

They are zero outside of the range 0 ≤ 𝑝, 𝑞 ≤ 𝑛. (In principle, these spaces could
have been defined using 𝜕 instead of 𝜕, but 𝜕 is preferred because it characterizes
holomorphic functions.)

Theorem 4.11 (Functoriality of Dolbeault Cohomology). If 𝐹 ∶ 𝑀 → 𝑁 is a

holomorphic map, then for each 𝑝 and 𝑞, the pullback 𝐹 ∗ ∶ ℰ 𝑝,𝑞 (𝑁) → ℰ 𝑝,𝑞 (𝑀)
descends to a linear map, also denoted by 𝐹 ∗ , from 𝐻 𝑝,𝑞 (𝑁) to 𝐻 𝑝,𝑞 (𝑀). It satisfies

(4.11) (Id𝑀 )∗ = Id ∶ 𝐻 𝑝,𝑞 (𝑀) → 𝐻 𝑝,𝑞 (𝑀),

(4.12) (𝐺 ∘ 𝐹 )∗ = 𝐹 ∗ ∘ 𝐺∗ if 𝐹 ∶ 𝑀 → 𝑁 and 𝐺 ∶ 𝑁 → 𝑃 are holomorphic.

Proof. The pullback satisfies 𝐹 ∗ ∘ 𝜕 = 𝜕 ∘ 𝐹 ∗ , so it defines a cochain map and

because they already hold when applied to (𝑝, 𝑞)-forms.

therefore descends to cohomology. Then (4.11) and (4.12) hold on cohomology
□

Dolbeault cohomology groups are biholomorphism invariants: If 𝐹 ∶ 𝑀 → 𝑁

Corollary 4.12 (Biholomorphism Invariance of Dolbeault Cohomology). The

is a biholomorphism, then for all 𝑝 and 𝑞, 𝐹 ∗ descends to an isomorphism from

𝐻 𝑝,𝑞 (𝑁) to 𝐻 𝑝,𝑞 (𝑀).

Proof. This follows from the fact that (𝐹 −1 )∗ is an inverse for 𝐹 ∗ . □

If 𝑀 is a complex manifold whose Dolbeault cohomology groups are finite-

dimensional (as we will see they always are when 𝑀 is compact), we define the
Hodge numbers of 𝑴 to be ℎ𝑝,𝑞 (𝑀) = dim 𝐻 𝑝,𝑞 (𝑀). These are similar to the
Betti numbers of a smooth manifold, 𝑏𝑘 (𝑀) = dim 𝐻dR 𝑘
(𝑀), where 𝐻dR𝑘
(𝑀) de-
notes the 𝑘th de Rham cohomology group. But unlike the Betti numbers, which are
topological invariants, the Hodge numbers in general depend on the holomorphic
structure of the manifold.
A Poincaré Lemma for the Dolbeault Operator 109

A Poincaré Lemma for the Dolbeault Operator

The Dolbeault cohomology groups provide answers to the question “which 𝜕-closed
forms are 𝜕-exact,” just as the de Rham cohomology groups answer “which 𝑑-closed
forms are 𝑑-exact.” An important feature of the de Rham groups is that 𝑑-closed

of the Poincaré lemma [LeeSM, Thm. 17.14], which says that every 𝑑-closed form
forms on smooth manifolds are always locally exact. (This is a direct consequence

on a star-shaped open subset of ℝ𝑁 is exact.) Thus the de Rham groups really

analogous fact is also true for 𝜕-closed forms.

are reflecting global properties of the manifold. The next theorem shows that an

Theorem 4.13 (The 𝝏-Poincaré Lemma). Suppose 𝑀 is a complex manifold and

𝜔 is a smooth (𝑝, 𝑞)-form on 𝑀 that satisfies 𝜕𝜔 = 0, with 𝑞 ≥ 1. Then in a
neighborhood of each point there is a smooth (𝑝, 𝑞 − 1)-form 𝜂 with 𝜕𝜂 = 𝜔.
Before beginning the proof, we need to establish the following analytic lemma.
Lemma 4.14 (The Inhomogeneous Cauchy–Riemann Equations). Suppose 𝑈
is an open subset of ℂ𝑛 and 𝑓 ∶ 𝑈 → ℂ is smooth. For each point 𝑎 ∈ 𝑈 and each
𝑘 ∈ {1, … , 𝑛}, there is a neighborhood of 𝑎 on which there exists a smooth solution
𝑔 to the equation
𝜕𝑔
= 𝑓.
𝜕𝑧𝑘
(4.13)

If 𝑓 is holomorphic with respect to one of the variables 𝑧𝑗 for 𝑗 ≠ 𝑘, then so is 𝑔.

Proof. To simplify the notation a bit, we will carry out the proof in the case 𝑘 = 1.
Given 𝑎 = (𝑎1 , … , 𝑎𝑛 ) ∈ 𝑈 , choose 𝑟 > 0 such that the closed polydisk 𝐷𝑛𝑟 (𝑎) is
contained in 𝑈 . We wish to define 𝑔 ∶ 𝐷𝑟𝑛 (𝑎) → ℂ by
1 𝑓 (𝑤, 𝑧2 , … , 𝑧𝑛 )
𝑔(𝑧1 , … , 𝑧𝑛 ) = 𝑑𝑤 ∧ 𝑑𝑤.
2𝜋𝑖 ∫ 𝑤 − 𝑧1
(4.14)
𝐷𝑟 (𝑎1 )

make sure that the integral makes sense. Note that 𝑑𝑤∧𝑑𝑤 is a constant multiple of
Because the integrand is not continuous on the domain of integration, we must first

𝑑𝑢 ∧ 𝑑𝑣 (where we write 𝑤 = 𝑢 + 𝑖𝑣), and |𝑓 (𝑤, 𝑧2 , … , 𝑧𝑛 )|/|𝑤 − 𝑧1 | ≤ 𝐶/|𝑤 − 𝑧1 |

for some constant 𝐶. The function 𝑤 ↦ 𝐶/|𝑤 − 𝑧1 | is an integrable function of 𝑤
on any compact subset of the plane (whether interpreted as a Lebesgue integral or

polar coordinates centered at 𝑧1 . Thus the integral makes sense for each 𝑧, and 𝑔 is
as an improper Riemann integral), as can be verified by expressing the integral in

Now we show that 𝑔 is smooth in all variables, is holomorphic in 𝑧𝑗 if 𝑓 is,

a well-defined function.

and satisfies (4.13). Given a complex number 𝑏1 ∈ 𝐷𝑟 (𝑎1 ), choose 𝜀 > 0 such that
𝐷𝜀 (𝑏1 ) ⊆ 𝐷𝑟 (𝑎1 ), and let 𝜑 ∶ ℂ → ℝ be a smooth bump function such that 𝜑 ≡ 1 on
𝐷𝜀/2 (𝑏1 ) and supp 𝜑 ⊆ 𝐷𝜀 (𝑏1 ). Let 𝑓1 (𝑧) = (1−𝜑(𝑧1 ))𝑓 (𝑧) and 𝑓2 (𝑧) = 𝜑(𝑧1 )𝑓 (𝑧),
so 𝑓 = 𝑓1 + 𝑓2 with 𝑓1 identically zero whenever 𝑧1 ∈ 𝐷𝜀/2 (𝑏1 ), and the function
110 4. The Dolbeault Complex

𝑤 ↦ 𝑓2 (𝑤, 𝑧2 , … , 𝑧𝑛 ) is supported in 𝐷𝜀 (𝑏1 ) for each (𝑧2 , … , 𝑧𝑛 ). Note also that

if 𝑓 is holomorphic with respect to one of the variables 𝑧𝑗 for 𝑗 ≥ 2, then so are 𝑓1
and 𝑓2 because 𝜑 depends only on 𝑧1 . Correspondingly, let 𝑔 = 𝑔1 + 𝑔2 , where 𝑔𝑖
is defined as in (4.14) but with 𝑓𝑖 in place of 𝑓 .
For 𝑔1 , the integrand is smooth in all variables and the domain of integration is
compact, so 𝑔1 is smooth in all variables. In particular, as long as |𝑧1 − 𝑏1 | < 𝜀/2,
we can differentiate under the integral sign to conclude
𝜕𝑔1
(𝑧) = 0 = 𝑓1 (𝑧).
𝜕𝑧1
Moreover, if 𝑓1 is holomorphic with respect to 𝑧𝑗 for some 𝑗 ≥ 2, then differentating
under the integral sign shows that 𝑔1 satisfies the Cauchy–Riemann equations with

Now consider 𝑔2 . Choose 𝑅 large enough that 𝐷𝑅 (𝑧1 ) ⊇ 𝐷𝜀 (𝑏1 ) for every
respect to that variable and thus is also holomorphic.

𝑧1 ∈ 𝐷𝜀 (𝑏1 ). Because the function 𝑤 ↦ 𝑓2 (𝑤, 𝑧2 , … , 𝑧𝑛 ) is smooth and compactly

supported in 𝐷𝜀 (𝑏1 ) for each (𝑧2 , … , 𝑧𝑛 ), for any given 𝑧1 ∈ 𝐷𝜀 (𝑏1 ) we might
as well compute 𝑔2 by integrating over the entire disk 𝐷𝑅 (𝑧1 ). Given such a 𝑧1 ,
we make the change of variables 𝑤 = 𝑧1 + 𝑟𝑒𝑖𝜃 , 𝑑𝑤 = 𝑒𝑖𝜃 (𝑑𝑟 + 𝑖𝑟 𝑑𝜃), 𝑑𝑤 =
𝑒−𝑖𝜃 (𝑑𝑟 − 𝑖𝑟 𝑑𝜃) to conclude

1 𝑓2 (𝑤, 𝑧2 , … , 𝑧𝑛 )
𝑔2 (𝑧1 , … , 𝑧𝑛 ) = 𝑑𝑤 ∧ 𝑑𝑤
2𝜋𝑖 ∫𝐷𝑅 (𝑧1 ) 𝑤 − 𝑧1
1 𝑓2 (𝑧1 + 𝑟𝑒𝑖𝜃 , 𝑧2 , … , 𝑧𝑛 )
= (−2𝑖𝑟) 𝑑𝑟 ∧ 𝑑𝜃
2𝜋𝑖 ∫𝐷𝑅 (0) 𝑟𝑒𝑖𝜃
−1
2𝜋 𝑅
= 𝑓2 (𝑧1 + 𝑟𝑒𝑖𝜃 , 𝑧2 , … , 𝑧𝑛 )𝑒−𝑖𝜃 𝑑𝑟 𝑑𝜃.
𝜋 ∫0 ∫0

This holds for all 𝑧 ∈ 𝐷𝑅 𝑛

(𝑎) such that |𝑧1 − 𝑏1 | < 𝜀. This integrand is smooth in

that 𝑔2 is smooth, and if 𝑓2 is holomorphic with respect to 𝑧𝑗 , 𝑗 ≥ 2, then so is 𝑔2 .

all variables and the domain of integration is compact, so once again we conclude

To compute 𝜕𝑔2 /𝜕𝑧1 , we differentiate under the integral sign:

𝜕𝑔2 1 −1
2𝜋 𝑅 𝜕𝑓
(𝑧 , … , 𝑧 𝑛
) = (𝑧 + 𝑟𝑒𝑖𝜃 , 𝑧2 , … , 𝑧𝑛 )𝑒−𝑖𝜃 𝑑𝑟 𝑑𝜃.
2 1
𝜕𝑧1 𝜋 ∫0 ∫0 𝜕𝑧1

Now change variables back to 𝑤 = 𝑧1 + 𝑟𝑒𝑖𝜃 :

𝜕𝑔2 1 1 𝜕𝑓2 𝑑𝑤 ∧ 𝑑𝑤
(𝑧 , … , 𝑧𝑛 ) = (𝑤, 𝑧2 , … , 𝑧𝑛 ) .
𝜕𝑧 1 ∫
2𝜋𝑖 𝐷𝑅 (𝑧1 ) 𝜕𝑧 1 𝑤 − 𝑧1

Choose 𝛿 > 0 smaller than 𝑅. For 𝑤 in the compact annulus 𝐴𝑅,𝛿 = 𝐷𝑅 (𝑧1 ) ∖
𝐷𝛿 (𝑧1 ) and (𝑧2 , … , 𝑧𝑛 ) arbitrary, the integrand is smooth in 𝑤 and equal to 𝑑𝜂,
A Poincaré Lemma for the Dolbeault Operator 111

𝑓2 (𝑤, 𝑧2 , … , 𝑧𝑛 ) 𝑑𝑤
where
𝜂=− .
𝑤 − 𝑧1

𝜕𝑓2
By Stokes’s theorem, therefore,
1 𝑑𝑤 ∧ 𝑑𝑤 1
(𝑤, 𝑧2 , … , 𝑧𝑛 ) = 𝑑𝜂
2𝜋𝑖 ∫𝐴𝑅,𝛿 𝜕𝑧 1 𝑤 − 𝑧 1 2𝜋𝑖 ∫𝐴𝑅,𝛿
1 1
= 𝜂− 𝜂
2𝜋𝑖 ∫𝜕𝐷𝑅 (𝑧 )
1 2𝜋𝑖 ∫𝜕𝐷𝛿 (𝑧1 )
1
=− 𝜂,
2𝜋𝑖 ∫𝜕𝐷𝛿 (𝑧1 )

tation on the inner circle, and the last equality follows because we chose 𝑅 large
where the negative sign in the next-to-last equality results from the Stokes orien-

enough that 𝜂 is identically zero on the outer circle. To compute this last integral,
parametrize 𝜕𝐷𝛿 (𝑧1 ) by 𝑤 = 𝑧1 + 𝛿𝑒𝑖𝜃 for 𝜃 ∈ [0, 2𝜋], so the pullback of 𝜂 to the
parameter domain [0, 2𝜋] is −𝑓2 (𝑧1 + 𝛿𝑒𝑖𝜃 , 𝑧2 , … , 𝑧𝑛 )𝑖 𝑑𝜃. Therefore,
1 1
2𝜋
− 𝜂= 𝑓2 (𝑧1 + 𝛿𝑒𝑖𝜃 , 𝑧2 , … , 𝑧𝑛 ) 𝑑𝜃.
2𝜋𝑖 ∫𝜕𝐷𝛿 (𝑧 )
1 2𝜋 ∫0

As 𝛿 → 0, the integrand converges uniformly to 𝑓2 (𝑧1 , … , 𝑧𝑛 ), so the limit of the

∫0 𝑓2 (𝑧) 𝑑𝜃 = 𝑓2 (𝑧). Putting this all together, we conclude
1 2𝜋
integral is equal to 2𝜋

𝜕𝑔2 1
that

(𝑧) = 𝑑𝜂
𝜕𝑧1 2𝜋𝑖 ∫𝐷𝑟 (𝑎1 )
1 1
= lim 𝑑𝜂 = − lim 𝜂 = 𝑓2 (𝑧).
𝛿→0 2𝜋𝑖 ∫𝐴𝑅,𝛿 𝛿→0 2𝜋𝑖 ∫𝜕𝐷𝛿 (𝑧)

Combining this with the computation for 𝑓1 , we find that for 𝑧 ∈ 𝐷𝑟𝑛 (𝑎) such that
|𝑧1 − 𝑏1 | < 𝜀,
𝜕𝑓 𝜕𝑓 𝜕𝑓
(𝑧) = 1 (𝑧) + 2 (𝑧) = 𝑔1 (𝑧) + 𝑔2 (𝑧) = 𝑔(𝑧).
𝜕𝑧 𝜕𝑧 𝜕𝑧
Since 𝑏1 was arbitrary, the same formula holds on the entire polydisk 𝐷𝑟𝑛 (𝑎), thus
completing the proof. □

Proof of the 𝝏-Poincaré lemma. Let 𝑛 = dim 𝑀. This is purely a local question,
so for each point 𝑎 ∈ 𝑀, we can choose holomorphic coordinates (𝑧1 , … , 𝑧𝑛 ) cen-
tered at 𝑎 and use the coordinate map to consider 𝜔 as a 𝜕-closed (𝑝, 𝑞)-form on a
neighborhood 𝑈 of 0 in ℂ𝑛 . We make this assumption henceforth.
We begin with the special case 𝑝 = 0. When 𝑛 = 1, only the (0, 1) case is
nontrivial. Thus suppose 𝜔 = 𝑓 𝑑𝑧 is a smooth (0, 1)-form on a neighborhood
of 0 in ℂ. It is automatically 𝜕-closed because there are no nonzero (0, 2)-forms.
112 4. The Dolbeault Complex

Lemma 4.14 shows that there is a smooth function 𝑔 on a neighborhood of 0 such

that 𝜕𝑔/𝜕𝑧 = 𝑓 , which is equivalent to 𝜕𝑔 = 𝜔.
Now let 𝑛 > 1. Suppose 𝜔 is a smooth 𝜕-closed (0, 𝑞)-form on a neighborhood
𝑈 of 0 in ℂ𝑛 . For each 𝑘 ∈ {0, … , 𝑞}, let Λ𝑘 (𝑈 ) denote the smooth subbundle
0,𝑞

of Λ0,𝑞 (𝑈 ) spanned by 𝑞-fold wedge products involving only 𝑑𝑧1 , … , 𝑑𝑧𝑘 ; and let
ℰ𝑘 (𝑈 ) denote the space of smooth sections of Λ𝑘 (𝑈 ). We will prove by induction
0,𝑞 0,𝑞

on 𝑘 that whenever 𝜔 ∈ ℰ𝑘 (𝑈 ) and 𝜕𝜔 = 0, there exists 𝜂 ∈ ℰ 0,𝑞−1 (𝑈 ) such that

0,𝑞

𝜕𝜂 = 𝜔. When 𝑘 = 𝑛, this is the desired conclusion.

For 𝑘 = 0, ℰ𝑘 (𝑀) contains only 𝜔 = 0, so there is nothing to prove. Sup-
0,𝑞

pose 𝑘 ≥ 1 and the claim is true for ℰ𝑘−1 (𝑈 ), and let 𝜔 be an element of ℰ𝑘 (𝑈 )
0,𝑞 0,𝑞

satisfying 𝜕𝜔 = 0. By separating out the terms in 𝜔 that contain 𝑑𝑧𝑘 , we can write
𝜔 = 𝛼 + 𝑑𝑧𝑘 ∧ 𝛽, where 𝛼 ∈ ℰ𝑘−1 (𝑈 ) and 𝛽 ∈ ℰ𝑘−1 (𝑈 ). Write
0,𝑞 0,𝑞−1

′ ′
𝛼= 𝛼𝐼 𝑑𝑧𝑖1 ∧ ⋯ ∧ 𝑑𝑧𝑖𝑞 , 𝛽= 𝛽𝐽 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑞−1 ,
∑ ∑
𝐼 𝐽

where the only multi-indices 𝐼 and 𝐽 that occur have all indices less than 𝑘. Our
assumption is
0 = 𝜕𝜔 = 𝜕𝛼 − 𝑑𝑧𝑘 ∧ 𝜕𝛽
𝜕𝛼𝐼 𝑟
𝑛
′
= 𝑑𝑧 ∧ 𝑑𝑧𝑖1 ∧ ⋯ ∧ 𝑑𝑧𝑖𝑞
∑ ∑ 𝜕𝑧𝑟
𝐼 𝑟=1

𝜕𝛽𝐽 𝑠
𝑛
′
− 𝑑𝑧𝑘 ∧ 𝑑𝑧 ∧ 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑞−1 .
∑ ∑ 𝜕𝑧𝑠
𝐽 𝑠=1

Let 𝐽 = (𝑗1 , … , 𝑗𝑞−1 ) be any increasing multi-index of length 𝑞 − 1 with all indices
less than 𝑘, and let 𝑠 > 𝑘. When we evaluate the right-hand side of the above
expression on the basis vectors (𝜕/𝜕𝑧𝑘 , 𝜕/𝜕𝑧𝑠 , 𝜕/𝜕𝑧𝑗1 , … , 𝜕/𝜕𝑧𝑗𝑞−1 ), only one term
produces a nonzero result, and that result is −𝜕𝛽𝐽 /𝜕𝑧𝑠 . Thus each coefficient 𝛽𝐽 is
holomorphic in the variables 𝑧𝑘+1 , … , 𝑧𝑛 . Lemma 4.14 shows that for each such 𝐽 ,
in some neighborhood 𝑈0 of 0 there is a smooth function 𝛾𝐽 satisfying 𝜕𝛾𝐽 /𝜕𝑧𝑘 =
𝛽𝐽 , and 𝜕𝛾𝐽 /𝜕𝑧𝑠 = 0 for 𝑠 ≥ 𝑘 + 1. Let 𝛾 be the (0, 𝑞 − 1)-form
′
𝛾= 𝛾𝐽 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑞−1 ,
∑
𝐽

so that

𝜕𝛾𝐽 𝑠
𝑘
′
𝜕𝛾 = 𝑑𝑧 ∧ 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑞−1
∑ ∑ 𝜕𝑧𝑠
𝐽 𝑠=1

= 𝑑𝑧𝑘 ∧ 𝛽 mod ℰ𝑘−1 (𝑈0 ).

0,𝑞
A Poincaré Lemma for the Dolbeault Operator 113

Therefore, 𝜔 − 𝜕𝛾 ∈ ℰ𝑘−1 (𝑈0 ). Since 𝜔 − 𝜕𝛾 is 𝜕-closed, the inductive hypothesis

0,𝑞

guarantees that there is a smooth (0, 𝑞 − 1)-form 𝜎 such that 𝜕𝜎 = 𝜔 − 𝜕𝛾, so we

have 𝜕(𝜎 + 𝛾) = 𝜔. This completes the inductive step and thus proves the lemma
in the 𝑝 = 0 case.
Finally, let 𝑝 be arbitrary, and suppose 𝜔 is a 𝜕-closed (𝑝, 𝑞)-form. We can write
′ ′
𝜔= 𝜔𝐼𝐽 𝑑𝑧𝑖1 ∧ ⋯ ∧ 𝑑𝑧𝑖𝑝 ∧ 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑞 = 𝛼𝐼 ∧ 𝛽𝐼 ,
∑ ∑
𝐼,𝐽 𝐼

where

𝛼𝐼 = 𝑑𝑧𝑖1 ∧ ⋯ ∧ 𝑑𝑧𝑖𝑝 ,
′
𝛽𝐼 = 𝜔𝐼𝐽 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑞 .
∑
𝐽

Note that 𝜕𝛼𝐼 = 0 because 𝑑𝛼𝐼 = 0, so

′
0 = 𝜕𝜔 = (−1)𝑝 𝛼𝐼 ∧ 𝜕𝛽𝐼 .
∑
𝐼

Choose a particular multi-index 𝐼 = (𝑖1 , … , 𝑖𝑝 ), and take the interior product of

both sides of this equation with the vector fields 𝜕/𝜕𝑧𝑖1 , … , 𝜕/𝜕𝑧𝑖𝑝 in turn, using
the fact that interior multiplication is an antiderivation (meaning that 𝑣 ⅃ (𝜔 ∧ 𝜂) =
(𝑣 ⅃ 𝜔)∧𝜂+(−1)𝑘 𝜔 ∧ (𝑣 ⅃ 𝜂) when 𝜔 and 𝜂 are forms of degrees 𝑘 and 𝑙, respectively
[LeeSM, Lemma 14.13]). This yields
𝜕 𝜕
0= ⅃⋯⅃ ⅃ 𝜕𝜔 = (−1)𝑝 𝜕𝛽𝐼 ,
𝜕𝑧𝑖𝑝 𝜕𝑧𝑖1
so 𝜕𝛽𝐼 = 0 for each 𝐼. Therefore, by the 𝑝 = 0 case proved above, for each 𝐼 there
is a (0, 𝑞 − 1)-form 𝜎𝐼 on a neighborhood of 0 such that 𝜕𝜎𝐼 = 𝛽𝐼 , and it follows
that 𝜕 ( ∑′𝐼 𝛼𝐼 ∧ 𝜎𝐼 ) = 𝜔. □

Corollary 4.15 (Local 𝝏𝝏-Lemma). Suppose 𝜃 is a smooth, closed (𝑝, 𝑞)-form on

a complex manifold 𝑀, with 𝑝 and 𝑞 both positive. In a neighborhood of each point
of 𝑀, there exists a smooth (𝑝 − 1, 𝑞 − 1)-form 𝛼 such that 𝜃 = 𝑖𝜕𝜕𝛼. If 𝜃 is a real
(𝑝, 𝑝)-form, then 𝛼 can be chosen to be real.

Proof. Because 𝜃 is a closed (𝑝 + 𝑞)-form, in a neighborhood of each point there

is a complex (𝑝 + 𝑞 − 1)-form 𝜂 such that 𝑑𝜂 = 𝜃 by the ordinary Poincaré lemma.

the real and imaginary parts.) Since the only parts of 𝜂 that can contribute to the
(The Poincaré lemma applies to complex-valued forms by applying it separately to

(𝑝, 𝑞)-part of 𝑑𝜂 are the (𝑝, 𝑞 − 1) and (𝑝 − 1, 𝑞) parts, we may as well assume that
𝜂 decomposes as 𝜂 = 𝜂 (𝑝,𝑞−1) + 𝜂 (𝑝−1,𝑞) .
114 4. The Dolbeault Complex

Using the fact that 𝑑 = 𝜕 + 𝜕, we can decompose the equation 𝑑𝜂 = 𝜃 as

follows:
𝜕𝜂 (𝑝−1,𝑞) = 0 ((𝑝 − 1, 𝑞 + 1) part),
𝜕𝜂 (𝑝,𝑞−1)
+ 𝜕𝜂 (𝑝−1,𝑞)
=𝜃 ((𝑝, 𝑞) part),
𝜕𝜂 (𝑝,𝑞−1)
=0 ((𝑝 + 1, 𝑞 − 1) part).
Now apply the 𝜕-Poincaré lemma to conclude that (after shrinking the neighborhood
if necessary) there exists a (𝑝 − 1, 𝑞 − 1)-form 𝛽 such that 𝜕𝛽 = 𝜂 (𝑝−1,𝑞) . Similarly,
since 𝜂 (𝑝,𝑞−1) is 𝜕-closed, there exists a (𝑞 − 1, 𝑝 − 1)-form 𝛾 such that 𝜕𝛾 = 𝜂 (𝑝,𝑞−1) .
Set 𝛼 = 𝑖𝛾 − 𝑖𝛽, which is a (𝑝 − 1, 𝑞 − 1)-form. Using the fact that 𝜕𝜕 = −𝜕𝜕,
we compute
𝑖𝜕𝜕𝛼 = 𝑖𝜕𝜕(𝑖𝛾) − 𝑖𝜕𝜕(𝑖𝛽)
= 𝜕(𝜕𝛾) + 𝜕(𝜕𝛽)
= 𝜕𝜂 (𝑝,𝑞−1) + 𝜕𝜂 (𝑝−1,𝑞)
= 𝜃.

In case 𝜃 is a real (𝑝, 𝑝)-form, we can choose 𝜂 to be real, which means 𝜂 (𝑝,𝑝−1) =
𝜂 (𝑝−1,𝑝) ; and then we can choose 𝛾 = 𝛽, so that 𝛼 = 𝑖𝛽 − 𝑖𝛽 is real. □

Bundle-Valued Forms
For holomorphic vector bundles, there is a generalized Dolbeault complex built out

We begin with a smooth manifold 𝑀 and a smooth complex vector bundle

of “bundle-valued differential forms.” Here is how that works.

𝐸 → 𝑀. Let End(𝐸) → 𝑀 be its endomorphism bundle, which is canonically

isomorphic to 𝐸 ⊗ 𝐸 ∗ . (See Example 3.24 for the holomorphic case; the argument
is exactly the same for smooth bundles.) For each nonnegative integer 𝑞, we define
the bundle of 𝑬-valued 𝒒-forms as the tensor product bundle Λℂ 𝑀 ⊗ 𝐸. We will
𝑞

use the notation ℰ 𝑞 (𝑀; 𝐸) to denote the space of smooth sections of Λℂ 𝑀 ⊗ 𝐸,

𝑞

so ℰ 0 (𝑀; 𝐸) is just the space of smooth sections of 𝐸 itself. Similarly, Λℂ 𝑀 ⊗

𝑞

End(𝐸) is the bundle of endomorphism-valued 𝒒-forms, and ℰ (𝑀; End(𝐸)) is its

𝑞

space of smooth sections.

morphism-valued forms. Suppose 𝐸 → 𝑀 is a smooth complex vector bundle. The

There are several natural wedge product operations on bundle-valued and endo-

forms and 𝐸-valued differential forms. For 𝛼 ∈ ℰ 𝑞 (𝑀) and 𝛽 ⊗ 𝜎 ∈ ℰ 𝑞 (𝑀; 𝐸),
simplest case is the wedge product between ordinary (scalar-valued) differential
′

𝛼 ∧ (𝛽 ⊗ 𝜎) = (𝛼 ∧ 𝛽) ⊗ 𝜎 ∈ ℰ 𝑞+𝑞 (𝑀; 𝐸),

we define
′

depends bilinearly on 𝛽 and 𝜎, it follows from the characteristic property of tensor

and then extend linearly to sums of such tensor products. Because this expression
Bundle-Valued Forms 115

over ℂ or ℝ) that this is well defined. In terms of a local frame (𝑠𝑗 ) for 𝐸, we can
product spaces [LeeSM, Prop. 12.7] (which holds equally well for tensor products

write 𝛽 = 𝛽 𝑗 ⊗ 𝑠𝑗 for some scalar-valued forms 𝛽 𝑗 (using the summation conven-

tion), and then
(4.15) 𝛼 ∧ (𝛽 𝑗 ⊗ 𝑠 𝑗 ) = (𝛼 ∧ 𝛽 𝑗 ) ⊗ 𝑠 𝑗 .
In particular, if 𝛽 is an 𝐸-valued 0-form (i.e., a smooth section of 𝐸), then 𝛼 ∧ 𝛽 is
just the same as 𝛼 ⊗𝛽; and if 𝛼 is a 0-form (i.e., a smooth function), then 𝛼 ∧𝛽 = 𝛼𝛽.
There is also a wedge product operation between 𝐸-valued forms and 𝐸 ∗ -
valued forms, which produces scalar forms. We define a wedge operation
∧ ∶ ℰ 𝑞 (𝑀; 𝐸 ∗ ) × ℰ 𝑞 (𝑀; 𝐸) → ℰ 𝑞+𝑞 (𝑀)
′ ′

by setting
(𝛾 ⊗ 𝜑) ∧ (𝛽 ⊗ 𝜎) = 𝜑(𝜎)(𝛾 ∧ 𝛽) ∈ ℰ 𝑞+𝑞 (𝑀)
′
(4.16)
for 𝛾 ⊗ 𝜑 ∈ ℰ 𝑞 (𝑀; 𝐸 ∗ ) and 𝛽 ⊗ 𝜎 ∈ ℰ 𝑞 (𝑀; 𝐸), and extending bilinearly. This
′

yields the following expression in terms of a local frame (𝑠𝑗 ) for 𝐸 and its dual
frame (𝜀𝑘 ) for 𝐸 ∗ :
( 𝛾𝑘 ⊗ 𝜀 ) ∧ ( 𝛽 ⊗ 𝑠 𝑗 ) = 𝛾 𝑗 ∧ 𝛽 .
𝑘 𝑗 𝑗
(4.17)

For 𝛼 ⊗𝐴 ∈ ℰ 𝑞 (𝑀; End(𝐸)), 𝛽 ⊗𝐵 ∈ ℰ 𝑞 (𝑀; End(𝐸)), and 𝛾 ⊗𝜎 ∈ ℰ 𝑞 (𝑀; 𝐸),

Especially important are wedge products with endomorphism-valued forms.
′ ″

we define
(𝛼 ⊗ 𝐴) ∧ (𝛽 ⊗ 𝐵) = (𝛼 ∧ 𝛽) ⊗ (𝐴 ∘ 𝐵) ∈ ℰ 𝑞+𝑞 (𝑀; End(𝐸)),
′

(𝛼 ⊗ 𝐴) ∧ (𝛾 ⊗ 𝜎) = (𝛼 ∧ 𝛾) ⊗ 𝐴𝜎 ∈ ℰ 𝑞+𝑞 (𝑀; 𝐸),

′

and extend bilinearly. To see how to compute these locally, let (𝑠𝑗 ) be a local
frame for 𝐸 and (𝜀𝑘 ) the dual frame for 𝐸 ∗ . Because of the canonical isomor-
phism End(𝐸) ≅ 𝐸 ⊗ 𝐸 ∗ , each section 𝜔 ∈ ℰ 𝑞 (End(𝐸)) can be expressed locally

𝜔 = 𝜔 𝑘 ⊗ 𝑠𝑗 ⊗ 𝜀 𝑘 ,
𝑗
in the form

for a uniquely determined matrix (𝜔𝑘 ) of ordinary 𝑞-forms. The tensor product
𝑗

𝑠𝑗 ⊗ 𝜀𝑘 represents the endomorphism of 𝐸 whose action on a basis element 𝑠𝑖 is

(𝑠𝑗 ⊗ 𝜀 )(𝑠𝑖 ) = 𝛿𝑖 𝑠𝑗 , so the wedge product defined above satisfies
𝑘 𝑘

𝜔 ∧ 𝜂 = (𝜔𝑘 ⊗ 𝑠𝑗 ⊗ 𝜀𝑘 ) ∧ (𝜂𝑚𝑙 ⊗ 𝑠𝑙 ⊗ 𝜀𝑚 )
𝑗

= (𝜔𝑘 ∧ 𝜂𝑚𝑙 ) ⊗ (𝛿𝑙𝑘 𝑠𝑗 ⊗ 𝜀𝑚 )

𝑗
(4.18)
= (𝜔𝑘 ∧ 𝜂𝑚𝑘 ) ⊗ (𝑠𝑗 ⊗ 𝜀𝑚 ).
𝑗

In other words, the matrix of forms representing 𝜔 ∧ 𝜂 is the matrix product of the
ones representing 𝜔 and 𝜂, with individual entries combined via the wedge product.
116 4. The Dolbeault Complex

(In an expression like 𝜔𝑘 , we always interpret the upper index as the row number
𝑗

and the lower index as the column number.) Similarly, if 𝛾 = 𝛾 𝑚 ⊗ 𝑠𝑚 is the local
expression for an element of ℰ 𝑞 (𝑀; 𝐸), then
″

𝜔 ∧ 𝛾 = (𝜔𝑘 ∧ 𝛾 𝑘 ) ⊗ 𝑠𝑗 .
𝑗
(4.19)

Now suppose 𝑀 is a complex manifold. For each 𝑝, 𝑞, we define the bundle of

𝑬-valued (𝒑, 𝒒)-forms as the tensor product bundle Λ𝑝,𝑞 𝑀 ⊗ 𝐸, and ℰ 𝑝,𝑞 (𝑀; 𝐸)
denotes the space of smooth sections of Λ𝑝,𝑞 𝑀 ⊗ 𝐸.
When the bundle 𝐸 is holomorphic, something special happens.

pose 𝑀 is a complex manifold and 𝐸 → 𝑀 is a holomorphic vector bundle. There

Proposition 4.16 (Cauchy–Riemann Operator on Bundle-Valued Forms). Sup-

are operators 𝜕 𝐸 ∶ ℰ 𝑝,𝑞 (𝑀; 𝐸) → ℰ 𝑝,𝑞+1 (𝑀; 𝐸) satisfying the following proper-
ties:
(i) For 𝜎 ∈ ℰ 0,0 (𝑀; 𝐸) = Γ(𝐸), 𝜕 𝐸 𝜎 = 0 if and only if 𝜎 is a holomorphic
section.
(ii) For 𝛼 ∈ ℰ 𝑝,𝑞 (𝑀) and 𝛽 ∈ ℰ 𝑝 ,𝑞 ′
(𝑀; 𝐸),
′

𝜕 𝐸 (𝛼 ∧ 𝛽) = 𝜕𝛼 ∧ 𝛽 + (−1)𝑝+𝑞 𝛼 ∧ 𝜕 𝐸 𝛽.

(iii) For 𝛾 ∈ ℰ 𝑝,𝑞 (𝑀; 𝐸 ∗ ) and 𝛽 ∈ ℰ 𝑝 ,𝑞 ′

(𝑀; 𝐸),
′

𝜕(𝛾 ∧ 𝛽) = 𝜕 𝐸 ∗ 𝛾 ∧ 𝛽 + (−1)𝑝+𝑞 𝛾 ∧ 𝜕 𝐸 𝛽.

(iv) 𝜕 𝐸 ∘ 𝜕 𝐸 = 0.
(v) If 𝛼 ∈ ℰ 𝑝,𝑞 (𝑀; 𝐸) satisfies 𝜕 𝐸 𝛼 = 0, then in a neighborhood of each
point there exists 𝛽 ∈ ℰ 𝑝,𝑞−1 (𝑀; 𝐸) such that 𝜕 𝐸 𝛽 = 𝛼.

Proof. Suppose 𝜎 ∈ ℰ 𝑝,𝑞 (𝑀; 𝐸). In any open set 𝑈 ⊆ 𝑀 over which there is a
holomorphic local frame (𝑠𝑗 ) for 𝐸, we can write 𝜎|𝑈 = 𝜎 𝑗 ⊗ 𝑠𝑗 for scalar-valued
forms 𝜎𝑗 . We wish to define 𝜕 𝐸 𝜎 by setting

(4.20) 𝜕 𝐸 𝜎 |𝑈 = (𝜕𝜎 𝑗 ) ⊗ 𝑠𝑗 .

choice of holomorphic local frame. If (𝑠𝑘̃ ) is another holomorphic local frame, then
To ensure this makes sense globally, we need to check that it is independent of the

where the domains overlap we can write 𝑠𝑘̃ = 𝜏𝑘 𝑠𝑗 for some holomorphic functions
𝑗

𝜏𝑘 . Then 𝜎 = 𝜎
̃𝑘 𝑠𝑘̃ with 𝜏𝑘 𝜎
̃ = 𝜎 𝑗 . Because 𝜕𝜏𝑘 ≡ 0, we have
𝑗 𝑗 𝑘 𝑗

(𝜕𝜎 ) ⊗ 𝑠𝑗 = 𝜕 (𝜏𝑘 𝜎 𝜎 𝑘 ) ⊗ (𝜏𝑘 𝑠𝑗 ) = (𝜕̃

̃𝑘 ) ⊗ 𝑠𝑗 = (𝜕̃ 𝜎 𝑘 ) ⊗ 𝑠𝑘̃ .
𝑗 𝑗 𝑗

This proves that 𝜕 𝐸 is well defined.

Bundle-Valued Forms 117

To prove (i), suppose 𝜎 is a smooth section of 𝐸. In terms of any local holo-

morphic frame (𝑠𝑗 ), we can write 𝜎 = 𝑓 𝑗 𝑠𝑗 for some complex-valued functions 𝑓 𝑗 .
If 𝜎 is holomorphic, then each 𝑓 𝑗 is holomorphic, and (4.20) shows that 𝜕 𝐸 𝜎 = 0.
Conversely, if 𝜕 𝐸 𝜎 = 0, then 0 = (𝜕𝑓 𝑗 ) ⊗ 𝑠𝑗 ; since the sections 𝑠𝑗 are linearly
independent at each point, this shows that 𝜕𝑓 𝑗 = 0 for each 𝑗, so 𝜎 is holomorphic.
Now (ii) follows by writing 𝛽 ∈ ℰ 𝑝 (𝑀; 𝐸) locally as 𝛽 𝑗 ⊗ 𝑠𝑗 , so that
′ ,𝑞 ′

𝜕 𝐸 (𝛼 ∧ 𝛽) = 𝜕 𝐸 ((𝛼 ∧ 𝛽 𝑗 ) ⊗ 𝑠𝑗 )
= (𝜕𝛼 ∧ 𝛽 𝑗 + (−1)𝑝+𝑞 𝛼 ∧ 𝜕𝛽 𝑗 ) ⊗ 𝑠𝑗
= 𝜕𝛼 ∧ 𝛽 + (−1)𝑝+𝑞 𝛼 ∧ 𝜕 𝐸 𝛽.

To prove (iii), write 𝛽 locally as above, and write 𝛾 = 𝛾𝑘 ⊗ 𝜀𝑘 , where (𝜀𝑘 )

denotes the local holomorphic frame for 𝐸 ∗ dual to (𝑠𝑗 ), so that 𝛾 ∧ 𝛽 = 𝛾𝑗 ∧ 𝛽 𝑗 .
We compute

𝜕 𝐸 ∗ 𝛾 ∧ 𝛽 + (−1)𝑝+𝑞 𝛾 ∧ 𝜕 𝐸 𝛽
= (𝜕𝛾𝑘 ⊗ 𝜀𝑘 ) ∧ (𝛽 𝑗 ⊗ 𝑠𝑗 ) + (−1)𝑝+𝑞 (𝛾𝑘 ⊗ 𝜀𝑘 ) ∧ (𝜕𝛽 𝑗 ⊗ 𝑠𝑗 )
= 𝜕𝛾𝑗 ∧ 𝛽 𝑗 + (−1)𝑝+𝑞 𝛾𝑗 ∧ 𝜕𝛽 𝑗
= 𝜕 (𝛾𝑗 ∧ 𝛽 𝑗 ) = 𝜕(𝛾 ∧ 𝛽).

To prove (iv), let 𝜎 ∈ ℰ 𝑝,𝑞 (𝑀; 𝐸), and let (𝑠𝑗 ) be a holomorphic local frame
for 𝐸 on 𝑈 ⊆ 𝑀. Writing 𝜎 = 𝜎 𝑗 ⊗ 𝑠𝑗 on 𝑈 , we conclude from (4.20) that

𝜕 𝐸 (𝜕 𝐸 𝜎 ) = 𝜕 𝐸 ((𝜕𝜎 𝑗 ) ⊗ 𝑠𝑗 ) = (𝜕𝜕𝜎 𝑗 ) ⊗ 𝑠𝑗 = 0.

Finally, to prove (v), suppose 𝛼 satisfies 𝜕 𝐸 𝛼 = 0. In terms of a holomorphic

local frame, we can write 𝛼 = 𝛼 𝑗 ⊗ 𝑠𝑗 for some scalar-valued (𝑝, 𝑞)-forms 𝛼 𝑗 sat-
isfying (𝜕𝛼 𝑗 ) ⊗ 𝑠𝑗 = 0. As above, the pointwise linear independence of the 𝑠𝑗 ’s
implies 𝜕𝛼 𝑗 = 0 for each 𝑗, and then the 𝜕-Poincaré lemma yields (𝑝, 𝑞−1)-forms 𝛽 𝑗
in a neighborhood of each point such that 𝜕𝛽 𝑗 = 𝛼 𝑗 and thus 𝜕 𝐸 (𝛽 𝑗 ⊗ 𝑠𝑗 ) = 𝛼. □

Note that in general, there is no natural 𝜕 operator on holomorphic vector bun-

dles. What makes it possible to define 𝜕 𝐸 is the fact that the transition functions

𝜕 operator.
relating different holomorphic frames are holomorphic and thus killed by the scalar

cohomology groups with coefficients in 𝑬 as the vector spaces

Thanks to part (iv) of the preceding proposition, we can define the Dolbeault

Ker (𝜕 𝐸 ∶ ℰ 𝑝,𝑞 (𝑀; 𝐸) → ℰ 𝑝,𝑞+1 (𝑀; 𝐸))

𝐻 𝑝,𝑞 (𝑀; 𝐸) = .
Im (𝜕 𝐸 ∶ ℰ 𝑝,𝑞−1 (𝑀; 𝐸) → ℰ 𝑝,𝑞 (𝑀; 𝐸))
(4.21)
118 4. The Dolbeault Complex

Λ𝑝,0 𝑀. If (𝑧𝑗 ) and (𝑧𝑗̃ ) are overlapping holomorphic coordinate charts for 𝑀, then
A particularly important special case of this construction applies to the bundles

the local coordinate frames for Λ𝑝,0 𝑀 are related by

𝑛
𝜕 𝑧𝑗̃ 1 𝜕 𝑧𝑗̃ 𝑝 𝑖1
𝑑 𝑧̃ ∧ ⋯ ∧ 𝑑 𝑧̃ =
𝑗1 𝑗𝑝
⋯ 𝑑𝑧 ∧ ⋯ ∧ 𝑑𝑧𝑖𝑝 ,
∑ 𝜕𝑧𝑖1 𝜕𝑧 𝑖𝑝
𝑖1 ,…,𝑖𝑝 =1

so the transition functions between the two frames are holomorphic. Thus Λ𝑝,0 𝑀

bundle-valued 𝜕-operator can be identified with the ordinary 𝜕 in this case. (Note
has the structure of a holomorphic vector bundle, and Problem 4-2 shows that the

that the bundles Λ𝑝,𝑞 𝑀 for 𝑞 ≠ 0 do not have natural holomorphic structures, how-
ever; they are just smooth bundles.) A section 𝛼 of Λ𝑝,0 𝑀 is holomorphic if and
only if 𝜕𝛼 = 0.
We denote the space of holomorphic (i.e., 𝜕-closed) sections of Λ𝑝,0 𝑀 by
Ω𝑝 (𝑀); these sections are called holomorphic 𝒑-forms. One particular case that
will be of special importance is the case 𝑝 = 𝑛, where 𝑛 = dim 𝑀. In that case
Λ𝑛,0 𝑀 is a holomorphic line bundle, called the canonical bundle of 𝑴. It is typi-
cally denoted by 𝐾𝑀 → 𝑀, or sometimes just 𝐾 → 𝑀 if it will not cause confu-
sion. Its dual, denoted by 𝐾𝑀
∗
or 𝐾 ∗ , is called the anticanonical bundle.
Proposition 4.17 (Canonical Bundle of Projective Space). Let 𝐾 → ℂℙ𝑛 be
the canonical bundle of ℂℙ𝑛 and let 𝐻 → ℂℙ𝑛 be its hyperplane bundle. Then
𝐾 ≅ 𝐻 −(𝑛+1) .

Proof. Problem 4-4. □

More generally, if 𝐸 → 𝑀 is a holomorphic vector bundle, each of the bundles

Λ𝑝,0 ⊗ 𝐸 is also holomorphic, and holomorphic sections of such a bundle are those

with holomorphic sections of 𝐸. We denote the space of holomorphic sections of

that can be written locally as a finite sum of tensor products of holomorphic forms

Λ𝑝,0 ⊗ 𝐸 by Ω𝑝 (𝑀; 𝐸).

Problems
4-1. Let 𝑀 be a complex manifold and let 𝜔 be a 2-form on 𝑀. Show that 𝜔 is
of type (1, 1) if and only if 𝜔(𝐽 𝑋, 𝐽 𝑌 ) = 𝜔(𝑋, 𝑌 ) for all vector fields 𝑋
and 𝑌 , where 𝐽 ∶ 𝑇 𝑀 → 𝑇 𝑀 is the canonical almost complex structure
on 𝑀.
4-2. On a complex manifold 𝑀, show that the complex vector bundles Λ𝑝,𝑞 𝑀
and Λ𝑝,0 𝑀 ⊗ Λ0,𝑞 𝑀 are isomorphic, and under this isomorphism the
operator 𝜕 𝐸 for 𝐸 = Λ𝑝,0 𝑀 corresponds to the standard 𝜕 operator.
Problems 119

4-3. Let (𝑀, 𝐽 ) be an almost complex manifold of real dimension 2𝑛, and de-
fine ℰ 𝑝,𝑞 (𝑀) to be the space of smooth complex-valued (𝑝+𝑞)-forms that
vanish if more than 𝑝 of their arguments are sections of 𝑇 ′ 𝑀 or more than
𝑞 are sections of 𝑇 ″ 𝑀. Show that the following are equivalent:
(a) 𝐽 is integrable.
(b) For every pair of integers 𝑝, 𝑞 ∈ {0, … , 𝑛}, the exterior derivative
operator 𝑑 maps ℰ 𝑝,𝑞 (𝑀) to ℰ 𝑝+1,𝑞 (𝑀) ⊕ ℰ 𝑝,𝑞+1 (𝑀).
(c) 𝑑 (ℰ 0,1 (𝑀)) ⊆ ℰ 1,1 (𝑀) ⊕ ℰ 0,2 (𝑀).

4-4. Prove Proposition 4.17 (canonical bundle of projective space).

4-5. Prove that there are no nontrivial global holomorphic 𝑝-forms on ℂℙ𝑛 for
𝑝 > 0. [Hint: Proceed by induction on 𝑝, using interior products with the
vector fields 𝑍𝑗 of Problem 3-5 to prove the inductive step.]

complex torus is constant. [Hint: Show that if 𝑇 = ℂ𝑚 /Λ is a complex

4-6. Prove that every holomorphic map from a complex projective space to a

torus, the holomorphic 1-forms 𝑑𝑧1 , … , 𝑑𝑧𝑚 on ℂ𝑚 descend to holomor-

phic forms 𝜁 1 , … , 𝜁 𝑛 on 𝑇 , and apply the result of Problem 4-5 to 𝐹 ∗ 𝜁 𝑗
for 𝐹 ∶ ℂℙ𝑛 → 𝑇 .]

4-7. Let 𝑀 be a complex manifold. A smooth real-valued function 𝑢 on 𝑀

is said to be pluriharmonic if in every holomorphic chart, 𝑢 is harmonic
(in the usual one-complex-variable sense) as a function of each complex
coordinate when the others are held fixed. Show that the following are

(a) 𝑢 is pluriharmonic.
equivalent.

(b) 𝜕𝜕𝑢 = 0.
(c) For every holomorphic embedding 𝑗 ∶ 𝔻 ↪ 𝑀 of the unit disk into
𝑀, 𝑢 ∘ 𝑗 is harmonic (in the usual sense) on 𝔻.
(d) In a neighborhood of each point of 𝑀, 𝑢 is the real part of a holo-
morphic function.

4-8. If 𝑆 ⊆ 𝑀 is a complex submanifold of a complex manifold 𝑀, the conor-

mal bundle of 𝑺 in 𝑴 is the holomorphic bundle 𝑁 ∗ 𝑆 → 𝑆 dual to
the holomorphic normal bundle (see Example 3.23). Show that 𝑁 ∗ 𝑆 is
canonically isomorphic to the holomorphic subbundle of Λ1,0 𝑀|𝑆 con-
sisting of (1, 0)-forms that vanish when applied to vectors tangent to 𝑆.

4-9. Suppose 𝑆 ⊆ 𝑀 is a closed complex hypersurface in a complex manifold

𝑀, and let 𝑁 ∗ 𝑆 → 𝑀 be its conormal bundle (Problem 4-8). Prove that
𝑁 ∗ 𝑆 ≅ 𝐿∗𝑆 |𝑆 , where 𝐿𝑆 → 𝑀 is the line bundle associated with 𝑆.
120 4. The Dolbeault Complex

4-10. Suppose 𝑆 is a closed complex hypersurface in a complex manifold 𝑀.

Let 𝐾𝑆 and 𝐾𝑀 denote the canonical bundles of 𝑆 and 𝑀, respectively,
and 𝐿𝑆 the line bundle on 𝑀 associated with the hypersurface 𝑆. Prove
the following adjunction formula:
𝐾𝑆 ≅ (𝐾𝑀 ⊗ 𝐿𝑆 )|𝑆 .
[Hint: Begin by covering a neighborhood of 𝑆 with holomorphic coordi-
nate charts in which the first coordinate vanishes simply on 𝑆.]
 10.1090/gsm/244/05

Chapter 5

Sheaves

There are many situations in differential geometry in which interesting geometric

objects can be produced locally (nonvanishing sections of bundles, orientations,
Riemannian metrics, logarithms of holomorphic functions, forms whose exterior
derivative is a given closed form, to name a few), leading to the question of whether
they can be pieced together to form a global object with the same properties. In
smooth manifold theory, this can sometimes be accomplished with a partition of
unity, as we do when proving the existence of Riemannian metrics. But there are
many such problems, such as deciding whether a closed form is exact or trying to
produce holomorphic objects of any type, in which multiplying by smooth bump
functions destroys the property we are attempting to preserve, so partitions of unity
are not useful. Sheaves were invented to systematize the process of going from local
objects to global ones in such situations. Their applications to complex manifolds
are plentiful and deep.
Sheaves were first introduced in the early 1940s by the French mathematician
Jean Leray (while he was a prisoner of war in Austria); the concept was then devel-
oped and refined by a number of mathematicians over the next two decades. The
article [Mil00] gives a good overview of the early history of the subject.

Definitions
The idea of a sheaf begins with a simple and familiar construction. Suppose 𝑀 is a
topological space. A presheaf on 𝑴 is an assignment to each open set 𝑈 ⊆ 𝑀 of
a set 𝒮 (𝑈 ), called the set of sections of 𝓢 over 𝑼 , along with a map 𝑟𝑈
𝑉 ∶ 𝒮 (𝑈 ) →
𝒮 (𝑉 ) called restriction whenever 𝑈 ⊇ 𝑉 , with the properties

(5.1) 𝑟𝑈
𝑈 = Id𝑈 for every open set 𝑈 ,
(5.2) 𝑟𝑉𝑊 ∘ 𝑟𝑈
𝑉 = 𝑟𝑊 whenever 𝑈 ⊇ 𝑉 ⊇ 𝑊 .
𝑈

121
122 5. Sheaves

structure. A presheaf 𝒮 is called a presheaf of abelian groups if each 𝒮 (𝑈 ) has

For our applications, we will require our presheaves to have some algebraic

an abelian group structure and the restriction maps are all homomorphisms; pre-
sheaves of rings and presheaves of real or complex vector spaces are defined sim-
ilarly. All of these are, in particular, presheaves of abelian groups (possibly with
extra structure), so sometimes we will concentrate on presheaves of abelian groups

tional structure. Some typical examples to keep in mind on a complex manifold 𝑀

and leave it to the reader to fill in the easy generalizations to presheaves with addi-

are the presheaf 𝒪 for which 𝒪(𝑈 ) is the ring of scalar-valued holomorphic func-
tions on 𝑈 , and the presheaf ℰ 𝑝,𝑞 for which ℰ 𝑝,𝑞 (𝑈 ) is the complex vector space
of smooth (𝑝, 𝑞)-forms on 𝑈 . (In order for these assignments to satisfy the defini-
tion of a presheaf, we must assign meanings to 𝒪(∅) and ℰ 𝑝,𝑞 (∅). You can check
that 𝒮 (∅) = {0} satisfies the definitions whenever 𝒮 is a presheaf of sets, abelian
groups, rings, or vector spaces, and we adopt this convention for all of our sheaves

Although there is no requirement that the elements of 𝒮 (𝑈 ) be maps of any

unless otherwise specified.)

to use the notation 𝑠|𝑉 for 𝑟𝑈𝑉 (𝑠); property (5.2) guarantees that its value does not
kind or that the “restriction maps” be actual restrictions of maps, it is convenient

depend on which larger open set 𝑈 we started with.

The definition of a presheaf can be expressed concisely in the language of cat-

a topological space 𝑀, let Top(𝑀) denote the category whose objects are open
egory theory (see, e.g., [LeeTM, pp. 209–214] or [Hat02, pp. 162–165]). Given

subsets of 𝑀 and whose morphisms are set inclusions. If C is any category what-
soever (typically sets, abelian groups, rings, or vector spaces), a presheaf on 𝑴
with values in C is a contravariant functor from Top(𝑀) to C.
Because the purpose of sheaf theory is to systematize the process of piecing
together local data to produce global results, our main objects of study will be pre-
sheaves with two additional properties, which guarantee that each section is locally
determined and that compatible local sections can be glued together to produce

ever {𝑈𝛼 }𝛼∈𝐴 is a collection of open subsets of 𝑀 and 𝑈 = ⋃𝛼 𝑈𝛼 :

global ones. A sheaf is a presheaf that satisfies the following two conditions when-

• LOCALITY PROPERTY: If 𝑠 and 𝑡 are elements of 𝒮 (𝑈 ) such that 𝑠|𝑈𝛼 = 𝑡|𝑈𝛼

for each 𝛼 ∈ 𝐴, then 𝑠 = 𝑡.

• GLUING PROPERTY: If we are given elements 𝑠𝛼 ∈ 𝒮 (𝑈𝛼 ) for each 𝛼 such

that 𝑠𝛼 |𝑈𝛼 ∩𝑈𝛽 = 𝑠𝛽 |𝑈𝛼 ∩𝑈𝛽 whenever 𝑈𝛼 ∩𝑈𝛽 ≠ ∅, then there is an element
𝑠 ∈ 𝒮 (𝑈 ) such that 𝑠𝛼 = 𝑠|𝑈𝛼 for every 𝛼 ∈ 𝐴.

► Exercise 5.1. Show that the following presheaves (with the restriction maps
given by actual restriction) satisfy the locality and gluing properties.
Definitions 123

(a) Given a topological, smooth, or holomorphic vector bundle 𝐸 → 𝑀, 𝒮 (𝑈 )

is the vector space of continuous, smooth, or holomorphic sections of 𝐸 over
𝑈 , respectively (with 𝒮 (∅) = {0} as explained above).
(b) Given any continuous map 𝜋 ∶ 𝑋 → 𝑀 between topological spaces, 𝒮 (𝑈 )
is the set of local sections of 𝝅 over 𝑈 , that is, continuous maps 𝜎 ∶ 𝑈 → 𝑋
such that 𝜋 ∘ 𝜎 = Id𝑈 . (In this case, 𝒮 (∅) = {∅}, because the empty map
∅ ⊆ ∅ × 𝑋 is the unique map from ∅ to any set.)
(c) With 𝑋, 𝑀, and 𝜋 as above, 𝒮̂(𝑈 ) is the set of rough local sections of 𝝅 over
𝑈 , that is, maps 𝜎 ∶ 𝑈 → 𝑋 that satisfy 𝜋 ∘ 𝜎 = Id𝑈 but are not necessarily
continuous.

Here are some general constructions we will use. Let 𝑀 be a topological space.
If 𝒮 is a sheaf on 𝑀 and 𝑉 ⊆ 𝑀 is an open subset, we define the restriction
of 𝓢 to 𝑽 , denoted by 𝒮 |𝑉 , to be the presheaf
(5.3) 𝒮 |𝑉 (𝑈 ) = 𝒮 (𝑈 ) for all open sets 𝑈 ⊆ 𝑉 ,
with the restriction maps inherited from 𝒮 . It follows easily from the definitions
that 𝒮 |𝑉 satisfies the gluing and locality properties and thus is a sheaf.
Given a sheaf 𝒮 on 𝑀, a subpresheaf 𝓣 of 𝓢 is a choice of subset (or sub-
group, subring, or vector subspace as appropriate) 𝒯 (𝑈 ) ⊆ 𝒮 (𝑈 ) for each open
set 𝑈 ⊆ 𝑀, such that 𝑟𝑈 𝑉 (𝒯 (𝑈 )) ⊆ 𝒯 (𝑉 ) whenever 𝑈 ⊇ 𝑉 . If 𝒯 satisfies the

subsheaf of 𝓢.
gluing and locality properties with the induced restriction maps, then it is called a

If 𝒮1 , … , 𝒮𝑘 are sheaves of abelian groups on 𝑀, their direct sum is the sheaf

𝒮1 ⊕ ⋯ ⊕ 𝒮𝑘 defined by (𝒮1 ⊕ ⋯ ⊕ 𝒮𝑘 )(𝑈 ) = 𝒮1 (𝑈 ) ⊕ ⋯ ⊕ 𝑆𝑘 (𝑈 ). It is
straightforward to check that this is a sheaf.
Examples of sheaves abound.
Example 5.2 (Sheaves). Let 𝑀 be a smooth manifold and 𝐸 → 𝑀 a smooth
complex vector bundle. The sheaves 𝒞 , 𝒞 ∗ , ℰ , ℰ ∗ , ℰ 𝑘 , 𝒵 𝑘 , ℰ (𝐸), ℰ 𝑘 (𝐸), 𝐺, and
𝐺𝑝 on 𝑀 are defined by the following assignments to each open set 𝑈 ⊆ 𝑀; in
each case, the restriction maps are the obvious ones unless otherwise specified.
(a) 𝒞 (𝑈 ) is the vector space of continuous complex-valued functions on 𝑈 .
(b) 𝒞 ∗ (𝑈 ) is the set of nowhere-vanishing continuous complex-valued func-
tions on 𝑈 , which is an abelian group under pointwise multiplication.
(c) ℰ (𝑈 ) is the vector space of smooth complex-valued functions on 𝑈 .
(d) ℰ ∗ (𝑈 ) is the abelian group of nowhere-vanishing smooth complex func-
tions on 𝑈 .
(e) ℰ 𝑘 (𝑈 ) is the vector space of smooth complex 𝑘-forms on 𝑈 .
(f) 𝒵 𝑘 (𝑈 ) is the vector space of smooth closed complex 𝑘-forms on 𝑈 .
(g) ℰ (𝐸)(𝑈 ) = ℰ (𝑈 ; 𝐸) is the vector space of smooth sections of 𝐸 over 𝑈 .
124 5. Sheaves

(h) ℰ 𝑘 (𝐸)(𝑈 ) = ℰ 𝑘 (𝑈 ; 𝐸) is the vector space of smooth 𝐸-valued complex

𝑘-forms on 𝑈 .
(i) 𝐺(𝑈 ), for any abelian group 𝐺, is the set of locally constant functions
from 𝑈 to 𝐺 (that is, functions that are constant in a neighborhood of
each point); it is a group under pointwise addition. The sheaf 𝐺 is called
the constant sheaf with coefficients in 𝐺. If 𝐺 is a ring or vector space,
then 𝐺 is a sheaf of rings or vector spaces, respectively.
(j) 𝐺𝑝 (𝑈 ), for a point 𝑝 ∈ 𝑀 and an abelian group 𝐺, is defined to be 𝐺 if
𝑝 ∈ 𝑈 , and {0} if not. The restriction map 𝑟𝑈
𝑉 is the identity if 𝑝 ∈ 𝑉 ⊆ 𝑈 ,
and otherwise it is the zero map. Any such sheaf is called a skyscraper
sheaf .
We have focused on sheaves of complex-valued functions and forms because they

real counterparts, which we will denote by a subscript ℝ when we have occasion to

are most useful in complex manifold theory; but some of these sheaves have obvious

use them—for example, ℰℝ is the sheaf of smooth real-valued functions on 𝑀.

In addition, when 𝑀 is a complex manifold, 𝐸 → 𝑀 is a holomorphic vector
bundle, and 𝑆 ⊆ 𝑀 is a closed subset, the sheaves 𝒪, 𝒪 ∗ , ℰ 𝑝,𝑞 , 𝒵 𝑝,𝑞 , Ω𝑝 , ℐ𝑆 , ℐ𝑆2 ,
𝒪(𝐸), ℰ 𝑝,𝑞 (𝐸), 𝒵 𝑝,𝑞 (𝐸), Ω𝑝 (𝐸), ℐ𝑆 (𝐸), and ℐ𝑆2 (𝐸) are defined as follows:
(k) 𝒪(𝑈 ) is the vector space of holomorphic functions from 𝑈 to ℂ.
(l) 𝒪 ∗ (𝑈 ) is the abelian group of nowhere-vanishing holomorphic functions
on 𝑈 .
(m) ℰ 𝑝,𝑞 (𝑈 ) is the space of smooth (𝑝, 𝑞)-forms on 𝑈 .
(n) 𝒵 𝑝,𝑞 (𝑈 ) is the space of smooth 𝜕-closed (𝑝, 𝑞)-forms on 𝑈 .
(o) Ω𝑝 (𝑈 ) is the space of holomorphic 𝑝-forms on 𝑈 (the same as 𝒵 𝑝,0 (𝑈 )).
(p) ℐ𝑆 (𝑈 ) is the set of holomorphic functions on 𝑈 that vanish on 𝑆 ∩ 𝑈 .
(q) ℐ𝑆2 (𝑈 ) is the vector space of holomorphic functions on 𝑈 that vanish to
second order on 𝑺 ∩ 𝑼 , which means that in a neighborhood 𝑉 of each
point they can be written as a finite sum ∑𝑗 𝑢𝑗 𝑣𝑗 where 𝑢𝑗 and 𝑣𝑗 are
holomorphic functions that vanish on 𝑆 ∩ 𝑉 .
(r) 𝒪(𝐸)(𝑈 ) = 𝒪(𝑈 ; 𝐸) is the space of holomorphic sections of 𝐸 over 𝑈 .
(s) ℰ 𝑝,𝑞 (𝐸)(𝑈 ) = ℰ 𝑝,𝑞 (𝑈 ; 𝐸) is the space of smooth 𝐸-valued (𝑝, 𝑞)-forms
on 𝑈 .
(t) 𝒵 𝑝,𝑞 (𝐸)(𝑈 ) = 𝒵 𝑝,𝑞 (𝑈 ; 𝐸) is the space of smooth 𝜕 𝐸 -closed 𝐸-valued
(𝑝, 𝑞)-forms on 𝑈 .
(u) Ω𝑝 (𝐸)(𝑈 ) = Ω𝑝 (𝑈 ; 𝐸) is the space of holomorphic 𝐸-valued 𝑝-forms on
𝑈.
(v) ℐ𝑆 (𝐸)(𝑈 ) = ℐ𝑆 (𝑈 ; 𝐸) is the set of holomorphic sections of 𝐸 over 𝑈
that vanish on 𝑆 ∩ 𝑈 .
Definitions 125

(w) ℐ𝑆2 (𝐸)(𝑈 ) = ℐ𝑆2 (𝑈 ; 𝐸) is the sheaf of holomorphic sections of 𝐸 over

𝑈 whose coefficient functions in any local holomorphic frame vanish to
second order on 𝑆 ∩ 𝑈 . //
It is worth noting that the sheaves 𝒞 , ℰ , and 𝒪 defined above also have the struc-
ture of sheaves of commutative rings under pointwise multiplication. The sheaf ℐ𝑆
is called the ideal sheaf of 𝑺, because ℐ𝑆 (𝑈 ) is an ideal in the ring 𝒪(𝑈 ).
On the other hand, it is not hard to come up with examples of presheaves that
are not sheaves.
Example 5.3 (Presheaves That Are Not Sheaves). Let 𝑀 be a manifold.
(a) ℛ(𝑈 ) is defined to be ℝ for every nonempty open set 𝑈 ⊆ 𝑀, and

𝑀 consists of only a single point, this presheaf is not a sheaf, because

all restriction maps to nonempty open sets are the identity map. Unless

it fails to satisfy the gluing property: whenever 𝑈𝛼 and 𝑈𝛽 are disjoint

nonempty open sets, we can choose distinct elements 𝑥𝛼 , 𝑥𝛽 ∈ ℝ and de-
fine “sections” 𝑥𝛼 ∈ ℛ(𝑈𝛼 ) and 𝑥𝛽 ∈ ℛ(𝑈𝛽 ), which vacuously satisfy
the compatibility condition because 𝑈𝛼 ∩ 𝑈𝛽 = ∅. But there is no section
𝑥 ∈ ℛ(𝑈𝛼 ∪ 𝑈𝛽 ) that restricts to these sections.
(b) ℛ0 (𝑈 ) = ℝ for every nonempty open set 𝑈 , but now all restriction maps
are zero maps. This presheaf fails to satisfy the locality property.
(c) ℬ(𝑈 ) is the set of bounded continuous complex-valued functions on 𝑈 .
If 𝑀 is noncompact, this presheaf fails to satisfy the gluing property. //

of rings. Thus suppose ℛ is a sheaf of commutative rings on 𝑀. (All our rings

For some purposes, it is important to consider sheaves of modules over a sheaf

are assumed to have a multiplicative identity, denoted by 1.) A sheaf 𝒮 on 𝑀 is

called a sheaf of 𝓡-modules if for each open set 𝑈 ⊆ 𝑀, 𝒮 (𝑈 ) can be given the
structure of a module over the ring ℛ(𝑈 ), and the restriction maps are compatible
with the module structure in the sense that for any open sets 𝑉 ⊆ 𝑈 ⊆ 𝑀 and
sections 𝑟 ∈ ℛ(𝑈 ), 𝑠 ∈ 𝒮 (𝑈 ), we have

(𝑟𝑠)|𝑉 = (𝑟|𝑉 )(𝑠|𝑉 ).

Example 5.4 (Sheaves of Modules).
(a) If 𝑀 is a smooth manifold and ℰ is its sheaf of smooth complex-valued
functions, the sheaves ℰ 𝑘 , ℰ (𝐸), and ℰ 𝑘 (𝐸) defined in Example 5.2 are
sheaves of ℰ -modules.
(b) If 𝑀 is a complex manifold, then ℰ 𝑝,𝑞 and ℰ 𝑝,𝑞 (𝐸) are likewise sheaves of
ℰ -modules on 𝑀; while Ω𝑝 , ℐ𝑆 , ℐ𝑆2 , 𝒪(𝐸), Ω𝑝 (𝐸), ℐ𝑆 (𝐸), and ℐ𝑆2 (𝐸)
are sheaves of 𝒪-modules, where 𝒪 is the sheaf of holomorphic functions
on 𝑀.
126 5. Sheaves

(c) Every sheaf 𝒮 of abelian groups on a topological space 𝑀 is a sheaf of

ℤ-modules, where ℤ is the constant sheaf on 𝑀 with coefficients in ℤ.
To verify this, suppose 𝑠 ∈ 𝒮 (𝑈 ) and 𝑘 ∈ ℤ(𝑈 ). Each point 𝑥 ∈ 𝑈 has
a neighborhood 𝑉𝑥 ⊆ 𝑈 on which 𝑘|𝑉𝑥 is constant, so (𝑘|𝑉𝑥 )(𝑠|𝑉𝑥 ) is an
element of the group 𝒮 (𝑉𝑥 ). By the gluing property, these local sections
patch together to determine a global section 𝑘𝑠 ∈ 𝒮 (𝑈 ). //

Sheaf Morphisms
Suppose 𝒮 and 𝒮 ′ are sheaves on the same space 𝑀, with restriction maps
𝑉 and 𝑟 𝑉 , respectively. A sheaf morphism from 𝓢 to 𝓢 is a collection
denoted by 𝑟𝑈 ′𝑈 ′

of maps 𝐹𝑈 ∶ 𝒮 (𝑈 ) → 𝒮 (𝑈 ), one for each open set 𝑈 ⊆ 𝑀, with the property that
′

they commute with restriction maps: whenever 𝑈 ⊇ 𝑉 , we have 𝑟′𝑈𝑉 ∘𝐹𝑈 = 𝐹𝑉 ∘𝑟𝑉 :
𝑈

𝐹𝑈
𝒮 (𝑈 ) / 𝒮 ′ (𝑈 )

(5.4) 𝑟𝑈
𝑉 𝑟′𝑈
𝑉

𝒮 (𝑉 ) / 𝒮 ′ (𝑉 ).
𝐹𝑉
 

If 𝒮 and 𝒮 ′ are sheaves of abelian groups, rings, or vector spaces, then we require
all the maps 𝐹𝑈 to be homomorphisms in the appropriate category. In addition,
if 𝒮 and 𝒮 ′ are both sheaves of modules over a sheaf ℛ of commutative rings, a
morphism 𝐹 ∶ 𝒮 → 𝒮 ′ is called an 𝓡-module morphism if each 𝐹𝑈 is an ℛ(𝑈 )-
module homomorphism. More generally, if 𝒮 and 𝒮 ′ are presheaves, then presheaf
morphisms from 𝒮 to 𝒮 ′ are defined in exactly the same way.

momorphisms 𝐹𝑈 by the same symbol 𝐹 . Using this notation together with the
Sometimes if there is no risk of confusion, we will just denote all of the ho-

shorthand notation for restrictions we introduced above, (5.4) can be written

(5.5) 𝐹 (𝑠|𝑉 ) = 𝐹 (𝑠)|𝑉 .

If 𝐹 ∶ 𝒮 → 𝒮 ′ and 𝐺 ∶ 𝒮 ′ → 𝒮 ″ are sheaf (or presheaf) morphisms, their

composition is the morphism 𝐺 ∘ 𝐹 ∶ 𝒮 → 𝒮 ″ defined by (𝐺 ∘ 𝐹 )𝑈 = 𝐺𝑈 ∘ 𝐹𝑈 for
each open set 𝑈 .
A sheaf morphism 𝐹 ∶ 𝒮 → 𝒮 ′ is called a sheaf isomorphism if there is a
sheaf morphism 𝐺 ∶ 𝒮 ′ → 𝒮 satisfying 𝐺 ∘ 𝐹 = Id𝒮 and 𝐹 ∘ 𝐺 = Id𝒮 ′ , or
equivalently if each 𝐹𝑈 is an isomorphism in the appropriate category; if there
exists such an isomorphism, we say 𝒮 and 𝒮 ′ are isomorphic sheaves, denoted by
𝒮 ≅ 𝒮 ′ . Presheaf isomorphisms and isomorphic presheaves are defined similarly.
The Étalé Space of a Presheaf 127

Example 5.5 (Sheaf Morphisms).

(a) If 𝒮 is a sheaf on a topological space 𝑀 and 𝒯 is a subsheaf of 𝒮 , there
is an inclusion morphism 𝑖 ∶ 𝒯 ↪ 𝒮 , defined by 𝑖𝑈 (𝜎) = 𝜎 for 𝜎 ∈
𝒯 (𝑈 ) ⊆ 𝒮 (𝑈 ).
(b) If 𝑀 is a smooth manifold, then for each nonnegative integer 𝑘 and each
open subset 𝑈 ⊆ 𝑀, the exterior derivative operator is a linear map
𝑑 ∶ ℰ 𝑘 (𝑈 ) → ℰ 𝑘+1 (𝑈 ). Because 𝑑 acts locally, it commutes with restric-
tions, and thus defines a sheaf morphism 𝑑 ∶ ℰ 𝑘 → ℰ 𝑘+1 . Similarly, if
𝑀 is a complex manifold, there are sheaf morphisms 𝜕 ∶ ℰ 𝑝,𝑞 → ℰ 𝑝,𝑞+1 .
(c) On a complex manifold 𝑀, let 𝒪 and 𝒪 ∗ be the sheaves of holomorphic

is a morphism 𝜀 ∶ 𝒪 → 𝒪 ∗ given by 𝜀𝑈 (𝑓 ) = 𝑒2𝜋𝑖𝑓 for 𝑓 ∈ 𝒪(𝑈 ) on an

functions and nonvanishing holomorphic functions, respectively. There

open subset 𝑈 ⊆ 𝑀. Because 𝜀𝑈 (𝑓 + 𝑔) = 𝜀𝑈 (𝑓 )𝜀𝑈 (𝑔), this is a sheaf

morphism. (Recall that 𝒪 ∗ is a sheaf of abelian groups under multiplica-

(d) Let 𝐺 and 𝐻 be abelian groups, and let 𝐺 and 𝐻 be the corresponding
tion.)

constant sheaves on some topological space 𝑀. Every group homomor-

phism 𝐹 ∶ 𝐺 → 𝐻 defines a sheaf morphism 𝐹 ∶ 𝐺 → 𝐻 by setting
𝐹 𝑈 (𝑓 ) = 𝐹 ∘ 𝑓 for 𝑓 ∈ 𝐺(𝑈 ). //
For any fixed topological space 𝑀, the sheaves on 𝑀 and sheaf morphisms
form a category. Similarly, there is a category of sheaves of abelian groups on 𝑀,

𝑀 with values in any category.

or similarly sheaves of rings, real or complex vector spaces, or indeed sheaves on

The Étalé Space of a Presheaf

In order to analyze the behavior of sheaf morphisms, we need to construct certain

struction, let us illustrate the principle in one concrete special case. Suppose 𝑀 is
sets called stalks associated with a presheaf or sheaf. Before doing the general con-

a complex manifold and 𝒪 is its sheaf of holomorphic functions. For each 𝑝 ∈ 𝑀,

we define an equivalence relation on the set of all holomorphic functions 𝑓 ∈ 𝒪(𝑈 )
for open sets 𝑈 that contain 𝑝, by saying 𝑓 ∈ 𝒪(𝑈 ) is equivalent to 𝑔 ∈ 𝒪(𝑉 ) if
𝑓 ≡ 𝑔 on some neighborhood of 𝑝. The equivalence class of 𝑓 ∈ 𝒪(𝑈 ) is called
the germ of 𝒇 at 𝒑 and denoted by [𝑓 ]𝑝 ; it is not usually necessary to specify a
particular open subset because the same germ is represented by the restriction of 𝑓
to any neighborhood of 𝑝, however small. The stalk of 𝓞 at 𝒑 is the vector space
𝒪𝑝 consisting of germs of all holomorphic functions at 𝑝. Addition and scalar mul-
tiplication of germs are defined by adding or multiplying any representatives that
are defined on the same open set. The same construction can be done with contin-
uous or smooth functions, or with continuous, smooth, or holomorphic sections of
a vector bundle.
128 5. Sheaves

algebraic construction. A directed set is a nonempty set 𝐼 with a binary relation “≤”
To define stalks for arbitrary presheaves we use instead the following general

that is reflexive and transitive, and such that any two elements have a common upper
bound. (In many examples, directed sets are actually partially ordered sets, but
they may not be, because the relation need not be antisymmetric; see the discussion
following the proof of Lemma 6.5 for an example.)

groups, rings, or vector spaces) is a family {𝐺𝛼 }𝛼∈𝐼 of objects indexed by a directed
A direct system of algebraic objects (which in our applications will be abelian

set 𝐼, together with a homomorphism 𝑓𝛼𝛽 ∶ 𝐺𝛼 → 𝐺𝛽 whenever 𝛼 ≤ 𝛽, satisfying

𝑓𝛼𝛼 = Id𝐺𝛼 for each 𝛼 and 𝑓𝛽𝛾 ∘ 𝑓𝛼𝛽 = 𝑓𝛼𝛾 whenever 𝛼 ≤ 𝛽 ≤ 𝛾.

all the 𝐺𝛼 ’s by saying 𝑔𝛼 ∈ 𝐺𝛼 is equivalent to 𝑔𝛽 ∈ 𝐺𝛽 if there is some 𝛾 ∈ 𝐼

Given a direct system, define an equivalence relation on the disjoint union of

with 𝛼 ≤ 𝛾 and 𝛽 ≤ 𝛾 such that 𝑓𝛼𝛾 (𝑔𝛼 ) = 𝑓𝛽𝛾 (𝑔𝛽 ) ∈ 𝐺𝛾 . The direct limit of the
direct system, denoted by lim 𝐺𝛼 , is the set of equivalence classes with addition, for
⟶
example, defined by
[𝑔𝛼 ] + [𝑔𝛽 ] = [𝑓𝛼𝛾 (𝑔𝛼 ) + 𝑓𝛽𝛾 (𝑔𝛽 )],
where 𝛾 is some upper bound for 𝛼 and 𝛽. This is well defined because all of the
maps 𝑓𝛼𝛽 are homomorphisms. Other operations such as ring multiplication or
scalar multiplication are defined similarly. For each object 𝐺𝛼 , there is a canonical
homomorphism from 𝐺𝛼 to the direct limit, obtained by sending 𝑔 ∈ 𝐺𝛼 to its

Suppose 𝒮 is a presheaf of abelian groups on a topological space 𝑀. For each

equivalence class.

𝑝 ∈ 𝑀, the collection of groups 𝒮 (𝑈 ) where 𝑈 ranges over all open sets containing
𝑝, together with the restriction maps, is a direct system with the relation 𝑈 ≤ 𝑉 if
𝑈 ⊇ 𝑉 . (The intersection of two open sets containing 𝑝 is a common upper bound.)
We define the stalk of 𝓢 at 𝒑, denoted by 𝒮𝑝 , to be the direct limit of this system.
The equivalence class of an element 𝑠 ∈ 𝒮 (𝑈 ) under this relation is denoted by
[𝑠]𝑝 ; by analogy with the construction for holomorphic functions described above,
we call [𝑠]𝑝 the germ of 𝒔 at 𝒑. You should convince yourself that this construction
applied to the sheaf 𝒪 yields germs of functions as we defined them above.

► Exercise 5.6. Suppose 𝑀 is a Hausdorff space and 𝐺𝑝 is a skyscraper sheaf

as defined in Example 5.2(j). Show that the stalk (𝐺𝑝 )𝑝 is isomorphic to 𝐺, while
for 𝑞 ≠ 𝑝, (𝐺𝑝 )𝑞 = 0. Give an example of a skyscraper sheaf on a non-Hausdorff
space for which this is not true.

Suppose 𝒮 and 𝒯 are presheaves of abelian groups on 𝑀. Every presheaf mor-

phism 𝐹 ∶ 𝒮 → 𝒯 induces natural homomorphisms on stalks as follows. Given
𝑝 ∈ 𝑀, we define the stalk homomorphism 𝐹𝑝 ∶ 𝒮𝑝 → 𝒯𝑝 by 𝐹𝑝 ([𝑠]𝑝 ) = [𝐹 (𝑠)]𝑝 ,
where 𝑠 ∈ 𝒮 (𝑈 ) is any representative of the germ [𝑠]𝑝 . To see that this is well
defined, suppose 𝑠 ∈ 𝒮 (𝑈 ) and 𝑠′ ∈ 𝒮 (𝑈 ′ ) represent the same germ at 𝑝. Then
The Étalé Space of a Presheaf 129

there is an open set 𝑊 ⊆ 𝑈 ∩ 𝑈 ′ containing 𝑝 such that 𝑠|𝑊 = 𝑠′ |𝑊 , and (5.5)

implies 𝐹 (𝑠)|𝑊 = 𝐹 (𝑠|𝑊 ) = 𝐹 (𝑠′ |𝑊 ) = 𝐹 (𝑠′ )|𝑊 , so 𝐹 (𝑠) and 𝐹 (𝑠′ ) represent the
same germ at 𝑝. Also, because 𝐹 and the restriction maps are homomorphisms on
each set 𝒮 (𝑈 ), it follows that 𝐹𝑝 is a homomorphism; and if 𝒮 is a sheaf of rings
or vector spaces, then 𝐹𝑝 is a homomorphism in the same category.
Using the concept of stalks, we will define a topological space canonically as-

for turning presheaves into sheaves. An étalé space1 over a topological space 𝑀
sociated with each presheaf, which will turn out to provide an important mechanism

is a topological space 𝐸 together with a local homeomorphism 𝜋 ∶ 𝐸 → 𝑀. The

preimage 𝐸𝑝 = 𝜋 −1 (𝑝) of a point 𝑝 ∈ 𝑀 is called the stalk of 𝑬 over 𝒑. An étalé
space is called an étalé space of abelian groups if each stalk has an abelian group

is a continuous map from 𝐸 to 𝐸, while group addition is continuous as a map from

structure and the group operations are continuous in the following sense: negation

𝐸 ×𝑀 𝐸 to 𝐸, where 𝐸 ×𝑀 𝐸 ⊆ 𝐸 ×𝐸 is the fiber product of 𝐸 with 𝐸 over 𝑀 (that

is, the subset {(𝑥1 , 𝑥2 ) ∈ 𝐸 × 𝐸 ∶ 𝜋(𝑥1 ) = 𝜋(𝑥2 )} with the subspace topology).
Étalé spaces of vector spaces or rings are defined analogously, with addition and

from ℝ × 𝐸 or ℂ × 𝐸 to 𝐸 when ℝ or ℂ is given the discrete topology.

multiplication continuous as above, and scalar multiplication continuous as a map

Theorem 5.7 (The Étalé Space of a Presheaf). Let 𝒮 be a presheaf over a topo-
logical space 𝑀, and let Et(𝒮 ) be the disjoint union of the stalks 𝒮𝑝 for all 𝑝 ∈ 𝑀,
with projection 𝜋 ∶ Et(𝒮 ) → 𝑀 defined by 𝜋([𝑠]𝑝 ) = 𝑝. For each open set 𝑈 ⊆ 𝑀
and each 𝑠 ∈ 𝒮 (𝑈 ), define a map 𝑠+ ∶ 𝑈 → Et(𝒮 ) by
(5.6) 𝑠+ (𝑝) = [𝑠]𝑝 .
Then Et(𝒮 ) has a unique topology such that 𝜋 is a local homeomorphism and each
𝑠+ is continuous. If 𝒮 is a presheaf of abelian groups, rings, or vector spaces, then
Et(𝒮 ) is an étalé space of objects in the same category.

Proof. We wish to take the collection of all subsets of the form 𝑠+ (𝑈 ) = {[𝑠]𝑝 ∶
𝑝 ∈ 𝑈 }, for open sets 𝑈 ⊆ 𝑀 and sections 𝑠 ∈ 𝒮 (𝑈 ), as a basis for a topology on
Et(𝒮 ). To see that it is a basis, we need to verify two things: (i) every point of Et(𝒮 )
is in some 𝑠+ (𝑈 ); and (ii) if two basis sets 𝑠+ (𝑈 ) and 𝑡+ (𝑉 ) intersect at a point [𝑤]𝑝 ,
then there is a basis set 𝑣+ (𝑊 ) such that [𝑤]𝑝 ∈ 𝑣+ (𝑊 ) ⊆ 𝑠+ (𝑈 ) ∩ 𝑡+ (𝑉 ). Property
(i) is immediate: every germ [𝑠]𝑝 is represented by some section 𝑠 ∈ 𝒮 (𝑈 ), and then
[𝑠]𝑝 is an element of the basis set 𝑠+ (𝑈 ). For (ii), assume [𝑤]𝑝 ∈ 𝑠+ (𝑈 ) ∩ 𝑡+ (𝑉 );
this means 𝑝 ∈ 𝑈 ∩ 𝑉 and the germs of 𝑤, 𝑠, and 𝑡 at 𝑝 are all equal. In other words,
there is some neighborhood 𝑊 ⊆ 𝑈 ∩ 𝑉 of 𝑝 such that 𝑤|𝑊 = 𝑠|𝑊 = 𝑡|𝑊 , and
then [𝑤]𝑝 ∈ 𝑤+ (𝑊 ) ⊆ 𝑠+ (𝑈 ) ∩ 𝑡+ (𝑉 ) as required.

1
The French word étalé is pronounced “ay-tah-LAY” and means “spread out.” There is another closely
related French word étale (“ay-TAHL”) without the second accent mark, meaning “slack,” which has a variety of
definitions in algebra and algebraic geometry; but étalé seems to have only this one mathematical meaning.
130 5. Sheaves

To see that each map 𝑠+ ∶ 𝑈 → Et(𝒮 ) is continuous, let 𝑡+ (𝑉 ) ⊆ Et(𝒮 ) be a

basis open set and observe that
(𝑠+ )−1 (𝑡+ (𝑉 )) = {𝑝 ∈ 𝑈 ∩ 𝑉 ∶ [𝑠]𝑝 = [𝑡]𝑝 }
= {𝑝 ∈ 𝑈 ∩ 𝑉 ∶ 𝑠|𝑊 = 𝑡|𝑊 for some neighborhood 𝑊 of 𝑝},
which is open in 𝑈 .
Next we show that 𝜋 is a local homeomorphism. It suffices to show that the
restriction of 𝜋 to each basis open set 𝑠+ (𝑈 ) is a homeomorphism onto 𝑈 . To see
that it is continuous, let 𝑉 ⊆ 𝑈 be open. Then 𝜋 −1 (𝑉 ) ∩ 𝑠+ (𝑈 ) is the set of germs
of 𝑠 at points of 𝑉 , or in other words exactly the basis set (𝑠|𝑉 )+ (𝑉 ). On the other
hand, 𝜋|𝑠+ (𝑈 ) ∶ 𝑠+ (𝑈 ) → 𝑈 has a continuous inverse given by the local section
𝑠+ ∶ 𝑈 → 𝑠+ (𝑈 ), so it is a homeomorphism onto its image.
To show that the topology is unique, suppose 𝐸 ̃ is the same set with another
topology such that 𝜋 is a local homeomorphism and each 𝑠+ is continuous. First we
̃ Thus let 𝜉 ∈ 𝑠+ (𝑈 ) be arbitrary, and set
will show that each set 𝑠+ (𝑈 ) is open in 𝐸.
̃
𝑝 = 𝜋(𝜉). Let 𝑊 be a neighborhood of 𝜉 in 𝐸 ̃ such that 𝜋|𝑊 ̃ is a homeomorphism
from 𝑊 ̃ to an open set 𝑊 ⊆ 𝑀, and let 𝑊0 = (𝑠+ )−1 (𝑊 ̃ ); because 𝑠+ is continuous
as a map to 𝐸,̃ 𝑊0 is an open neighborhood of 𝑝 contained in 𝑈 ∩ 𝑊 . Because 𝜋|𝑊 ̃
is a homeomorphism, 𝑊 ̃ 0 = (𝜋|𝑊̃ ) (𝑊0 ) is an open neighborhood of 𝜉 contained
−1

in 𝑊̃ . Also, the fact that 𝜋 ∘ 𝑠+ = Id𝑈 implies 𝜋(𝑊 ̃ 0 ) = 𝜋 ∘ 𝑠+ ∘ 𝜋(𝑊 ̃ 0 ), and

̃ ̃ ̃ ̃
injectivity of 𝜋 on 𝑊 then implies 𝑊 0 = 𝑠 ∘ 𝜋(𝑊 0 ). Thus 𝑊 0 is also contained
+

in 𝑠+ (𝑈 ). Since each point of 𝑠+ (𝑈 ) has a neighborhood contained in 𝑠+ (𝑈 ), we

̃
see that 𝑠+ (𝑈 ) is open in 𝐸.
Now both identity maps Et(𝒮 ) → 𝐸 ̃ and 𝐸 ̃ → Et(𝒮 ) are continuous, because
each is equal to the composition 𝑠 ∘𝜋 when restricted to one of the open sets 𝑠+ (𝑈 ).
+

If 𝒮 is a presheaf of abelian groups, rings, or vector spaces, then each stalk

Thus the two topologies are the same.

𝒮𝑝 inherits the appropriate algebraic structure from the direct limit operation as
described above. Let 𝑎 ∶ Et(𝒮 ) ×𝑀 Et(𝒮 ) → Et(𝒮 ) be addition. To see that it
is continuous, suppose 𝑠+ (𝑈 ) ⊆ Et(𝒮 ) is a basis open set. Then 𝑎−1 (𝑠+ (𝑈 )) is
the set of all pairs of the form ([𝑣]𝑝 , [𝑤]𝑝 ) where 𝑝 ∈ 𝑈 , 𝑣, 𝑤 ∈ 𝒮 (𝑌 ) for some
neighborhood 𝑌 of 𝑝 contained in 𝑈 , and 𝑣 + 𝑤 = 𝑠|𝑌 . That is to say,
𝑎−1 (𝑠+ (𝑈 )) = 𝑣+ (𝑌 ) × 𝑤+ (𝑌 ).
⋃
𝑝∈𝑌 ⊆𝑈
𝑣,𝑤∈𝑆(𝑌 )
𝑣+𝑤=𝑠|𝑌

As a union of open sets, this is open. Similar arguments show that the other alge-
braic operations are continuous in appropriate cases. □

Et(𝒮 ) in terms of sections of 𝒮 .

The next lemma gives a useful characterization of continuity of sections of
The Étalé Space of a Presheaf 131

Lemma 5.8. Let 𝒮 be a presheaf on a topological space 𝑀 and 𝜋 ∶ Et(𝒮 ) → 𝑀

its étalé space. For any subset 𝐵 ⊆ 𝑀, a rough section 𝜑 ∶ 𝐵 → Et(𝒮 ) (i.e.,
a map such that 𝜋 ∘ 𝜑 = Id𝐵 ) is continuous if and only if each point in 𝐵 has a
neighborhood 𝑉 in 𝑀 and a section 𝑠 ∈ 𝒮 (𝑉 ) such that 𝜑|𝑉 ∩𝐵 = 𝑠+ |𝑉 ∩𝐵 .

Proof. Suppose first that 𝜑 ∶ 𝐵 → Et(𝒮 ) is a continuous section. Given 𝑝 ∈ 𝐵,

the definition of Et(𝒮 ) shows that 𝜑(𝑝) = [𝑠]𝑝 for some section 𝑠 ∈ 𝒮 (𝑊 ) on
some neighborhood 𝑊 of 𝑝. Then 𝑠+ (𝑊 ) is a neighborhood of [𝑠]𝑝 in Et(𝒮 ), and
since 𝜑 is continuous, 𝑉 = 𝜑−1 (𝑠+ (𝑊 )) is a neighborhood of 𝑝 in 𝑈 . This means
𝜑(𝑞) = [𝑠]𝑞 for all 𝑞 ∈ 𝑉 ∩ 𝐵, or in other words 𝜑 agrees with the restriction of 𝑠+
to 𝑉 ∩ 𝐵.
Conversely, if 𝜑 satisfies the condition in statement of the lemma, then it is
continuous because it agrees with a continuous section 𝑠+ in a neighborhood of
each point. □

Theorem 5.9 (The Sheafification Functor). Suppose 𝑀 is a topological space

and 𝒮 is a presheaf on 𝑀. Then there is a sheaf 𝒮 + , called the sheafification
of 𝓢, together with a canonical presheaf morphism 𝜃𝒮 ∶ 𝒮 → 𝒮 + , satisfying the
following properties.
(a) If 𝒮 is a presheaf of abelian groups, rings, or vector spaces, then 𝒮 + is a
sheaf of objects in the same category.
(b) 𝒮 + satisfies the following universal property: If 𝒯 is any sheaf and
𝐹 ∶ 𝒮 → 𝒯 is any presheaf morphism, there is a unique sheaf morphism
𝐹 ∶ 𝒮 + → 𝒯 satisfying 𝐹 ∘ 𝜃𝒮 = 𝐹 :

𝐹 /
(5.7) 𝒮 =𝒯
𝜃𝒮
⑤ 𝐹
⑤
⑤

𝒮+
⑤


(c) 𝜃𝒮 is an isomorphism if and only if 𝒮 is a sheaf.

(d) If 𝒯 is any presheaf and 𝐹 ∶ 𝒮 → 𝒯 is a presheaf morphism, there is a
unique sheaf morphism 𝐹 + ∶ 𝒮 + → 𝒯 + satisfying 𝜃𝒯 ∘ 𝐹 = 𝐹 + ∘ 𝜃𝒮 :

𝐹
𝒮 /𝒯

(5.8) 𝜃𝒮 𝜃𝒯
𝒮 + ❴ ❴+❴/ 𝒯 + .
𝐹
 

It also satisfies (Id𝒮 )+ = Id𝒮 + and (𝐹 ∘ 𝐺)+ = 𝐹 + ∘ 𝐺+ .

132 5. Sheaves

Proof. Define the sheaf 𝒮 + on 𝑀 by letting 𝒮 + (𝑈 ) be the set of local sections of

Et(𝒮 ) over the open set 𝑈 ⊆ 𝑀, that is, continuous maps 𝜑 ∶ 𝑈 → Et(𝒮 ) such
that 𝜋 ∘ 𝜑 = Id𝑈 ; and letting the restriction maps be actual restrictions of sections.
This is a sheaf by the result of Exercise 5.1(b). The algebraic operations on 𝒮 + are
defined pointwise: for example, if 𝒮 is a presheaf of abelian groups, we define a
group structure on each set 𝒮 + (𝑈 ) by

(𝑠1 + 𝑠2 )(𝑝) = 𝑠1 (𝑝) + 𝑠2 (𝑝).

(If 𝑈 = ∅, we just interpret 𝒮 + (𝑈 ) = {∅} to be the trivial group.) The fact that the

Et(𝒮 ). The same argument applies to negation and to the other algebraic operations
group operations are continuous guarantees that the result is a continuous section of

in the case of a presheaf of rings or vector spaces. With these pointwise operations,
it is immediate that each restriction map is a homomorphism. This completes the

The presheaf morphism 𝜃𝒮 ∶ 𝒮 → 𝒮 + is defined by 𝜃𝒮 (𝑠) = 𝑠+ for 𝑠 ∈ 𝑆(𝑈 ).

proof of (a).

To prove (b), suppose 𝒯 is a sheaf and 𝐹 ∶ 𝒮 → 𝒯 is a presheaf morphism. We

define the sheaf morphism 𝐹 ∶ 𝒮 + → 𝒯 first by defining 𝐹 (𝑠+ ) = 𝐹 (𝑠) ∈ 𝒯 (𝑈 )
for any open set 𝑈 ∈ 𝑀 and 𝑠 ∈ 𝒮 (𝑈 ); this ensures that 𝐹 ∘ 𝜃𝒮 = 𝐹 . Then we
extend 𝐹 to act on an arbitrary section 𝜑 ∈ 𝒮 + (𝑈 ) as follows: Lemma 5.8 shows
that there is an open cover {𝑈𝛼 } of 𝑈 and sections 𝑠𝛼 ∈ 𝒮 (𝑈𝛼 ) such that 𝜑|𝑈𝛼 = 𝑠+ 𝛼
for each 𝛼. Let 𝐹 (𝜑) be the element 𝜏 ∈ 𝒯 (𝑈 ) such that 𝜏|𝑈𝛼 = 𝐹 (𝑠𝛼 ) for all 𝛼.
Such a section exists because 𝒯 satisfies the gluing property, and it is well defined
independently of the choice of open cover because 𝒯 satisfies the locality property.
Moreover, any other such map 𝐹 ′ would have to agree with this one because the
restriction of 𝐹 ′ (𝜑) to sufficiently small open sets has to agree with sections of the
form 𝐹 (𝑠+ ) = 𝐹 ∘ 𝜃𝒮 (𝑠). This completes the proof of (b).
To prove (c), note first that if 𝜃𝒮 is a presheaf isomorphism, then 𝒮 is certainly
a sheaf because 𝒮 + is. Conversely, assume 𝒮 is a sheaf. Applying (b) to Id𝒮 ∶ 𝒮 →
𝒮 , we conclude there is a sheaf morphism Id𝒮 ∶ 𝒮 + → 𝒮 such that Id𝒮 ∘ 𝜃𝒮 = Id𝒮 .
On the other hand, the same result applied to 𝜃𝒮 ∶ 𝒮 → 𝒮 + shows that there is a
unique morphism 𝑔 ∶ 𝒮 + → 𝒮 + satisfying 𝑔 ∘ 𝜃𝒮 = 𝜃𝒮 ; since both 𝑔 = Id𝒮 + and
𝑔 = 𝜃𝒮 ∘ Id𝒮 are such morphisms, we conclude that 𝜃𝒮 ∘ Id𝒮 = Id𝒮 + . Thus Id𝒮 is
an inverse for 𝜃𝒮 .
Finally, to prove (d), suppose 𝐹 ∶ 𝒮 → 𝒯 is a presheaf morphism. Define
𝐹 + ∶ 𝒮 + → 𝒯 + by 𝐹 + = 𝜃𝒯 ∘ 𝐹 , the morphism whose existence and uniqueness

The argument in the preceding paragraph showed that (Id𝒮 )+ = 𝜃𝒮 = Id𝒮 + . If

are guaranteed by the universal property of part (b); then (5.8) follows from (5.7).

𝐹 ∶ 𝒮 → 𝒯 and 𝐺 ∶ 𝒯 → 𝒰 are presheaf morphisms, then both ℎ1 = 𝐺+ ∘ 𝐹 +

and ℎ2 = (𝐺 ∘ 𝐹 )+ satisfy ℎ𝑗 ∘ 𝜃𝒮 = 𝜃𝒰 ∘ 𝐺 ∘ 𝐹 , so by the uniqueness part of
statement (b), they are equal. □
The Étalé Space of a Presheaf 133

If 𝜋 ∶ 𝐸 → 𝑀 and 𝜋 ′ ∶ 𝐸 ′ → 𝑀 are étalé spaces over 𝑀, an étalé space

morphism from 𝐸 to 𝐸 ′ is a continuous map 𝐹 ∶ 𝐸 → 𝐸 ′ such that 𝜋 ′ ∘ 𝐹 = 𝜋,
and such that 𝐹 restricts to a homomorphism on each stalk with respect to whatever
algebraic structure the stalks are endowed with. For each topological space 𝑀, we
can define a category Sh(𝑀) of sheaves (of abelian groups, say) on 𝑀 and sheaf
morphisms, and a category Et(𝑀) of étalé spaces of abelian groups on 𝑀 and étalé

Et ∶ Sh(𝑀) → Et(𝑀).
space morphisms. The construction of the étalé space of a sheaf defines a functor

exactly the same information. If C and D are categories, a functor ℱ ∶ C →

The next proposition shows, essentially, that sheaves and étalé spaces contain

ℱ (𝑐) for some object 𝑐 of C, and for every pair of objects 𝑐1 , 𝑐2 of C, the map
D is called an equivalence of categories if every object of D is isomorphic to

ℱ ∶ HomC (𝑐1 , 𝑐2 ) → HomD (ℱ (𝑐1 ), ℱ (𝑐2 )) is bijective.

Proposition 5.10. Let 𝑀 be a topological space, Sh(𝑀) the category of sheaves
of abelian groups on 𝑀, and Et(𝑀) the category of étalé spaces of abelian groups
over 𝑀. The étalé space functor Et ∶ Sh(𝑀) → Et(𝑀) is an equivalence of cate-
gories.

Proof. Problem 5-5. □

You will find that some authors define sheaves to be étalé spaces, and sheaf
morphisms to be étalé space morphisms. The preceding proposition shows that
one can translate back and forth between that definition and the one we have given
without any loss of information.

combinations of sheaves. We begin with quotient sheaves. Suppose 𝒮 is a sheaf

The sheafification functor has some important applications to defining natural

of abelian groups on 𝑀 and ℛ is a subsheaf of 𝒮 . In general, the presheaf 𝑈 ↦

𝒮 (𝑈 )/ℛ(𝑈 ) need not be a sheaf, as the next example illustrates.
Example 5.11. Suppose 𝑀 is a compact, connected complex manifold and 𝑝, 𝑞 are
distinct points in 𝑀. Let 𝒪 be the sheaf of holomorphic functions on 𝑀, and let
ℐ ↪ 𝒪 be the subsheaf of holomorphic functions that vanish at both 𝑝 and 𝑞 (the
ideal sheaf of {𝑝, 𝑞}). Let 𝑈1 = 𝑀 ∖ {𝑝} and 𝑈2 = 𝑀 ∖ {𝑞}. The constant function
1 on 𝑈1 determines a nontrivial element of the quotient space 𝒪(𝑈1 )/ℐ (𝑈1 ), and
the constant function 2 determines a nontrivial element of 𝒪(𝑈2 )/ℐ (𝑈2 ). Since
𝒪(𝑈1 ∩ 𝑈2 )/ℐ (𝑈1 ∩ 𝑈2 ) is the zero vector space, both elements restrict to zero in
this quotient space. But there is no element of 𝒪(𝑀)/ℐ (𝑀) that restricts to the
equivalence class of 1 on 𝑈1 and the equivalence class of 2 on 𝑈2 , because every
element of 𝒪(𝑀) is constant. Thus the presheaf 𝑈 ↦ 𝒪(𝑈 )/ℐ (𝑈 ) does not satisfy
the gluing property. //
To get around this problem, we define the quotient sheaf 𝒮 /ℛ to be the sheafi-
fication of the presheaf 𝑈 ↦ 𝒮 (𝑈 )/ℛ(𝑈 ).
134 5. Sheaves

ℛ is a sheaf of commutative rings on 𝑀, and 𝒮 and 𝒯 are both sheaves of ℛ-

Another use of sheafification is to define tensor products of sheaves. Suppose

modules. We can form the presheaf 𝑈 ↦ 𝒮 (𝑈 ) ⊗ℛ(𝑈 ) 𝒯 (𝑈 ), but once again this

product of 𝓢 and 𝓣 over 𝓡, denoted by 𝒮 ⊗ℛ 𝒯 , to be the sheafification of the

might not be a sheaf. (See Problem 5-9 for a counterexample.) We define the tensor

presheaf 𝑈 ↦ 𝒮 (𝑈 ) ⊗ℛ(𝑈 ) 𝒯 (𝑈 ). The next proposition illustrates the naturalness

of this concept.
Proposition 5.12. Suppose 𝑀 is a complex manifold and 𝐸, 𝐸 ′ → 𝑀 are holo-
morphic vector bundles over 𝑀. Then the tensor product sheaf 𝒪(𝐸) ⊗𝒪 𝒪(𝐸 ′ ) is
isomorphic to 𝒪(𝐸 ⊗ 𝐸 ′ ).

Proof. Let 𝑈 ⊆ 𝑀 be an open subset. An element 𝜎 ∈ 𝒪(𝑈 ; 𝐸) ⊗𝒪(𝑈 ) 𝒪(𝑈 ; 𝐸 ′ )

can be represented as a finite sum of abstract tensor products, ∑𝑗 𝜎𝑗 ⊗ 𝜎𝑗′ , where
𝜎𝑗 ∈ 𝒪(𝑈 ; 𝐸) and 𝜎𝑗′ ∈ 𝒪(𝑈 ; 𝐸 ′ ). We let 𝐹𝑈 (𝜎) be the holomorphic section
of the tensor product bundle 𝐸 ⊗ 𝐸 ′ over 𝑈 given by the same formula: 𝑥 ↦
∑𝑗 𝜎𝑗 (𝑥) ⊗ 𝜎𝑗′ (𝑥). Because this formula is bilinear over 𝒪(𝑈 ), this gives a well-
defined homomorphism

𝐹𝑈 ∶ 𝒪(𝑈 ; 𝐸) ⊗𝒪(𝑈 ) 𝒪(𝑈 ; 𝐸 ′ ) → 𝒪(𝑈 ; 𝐸 ⊗ 𝐸 ′ ),

and because 𝐹𝑉 is the restriction of 𝐹𝑈 whenever 𝑉 ⊆ 𝑈 , this defines a presheaf

morphism. By Theorem 5.9(b), it induces a sheaf morphism 𝐹 from 𝒪(𝐸)⊗𝒪 𝒪(𝐸 ′ )
to 𝒪(𝐸 ⊗ 𝐸 ′ ).
We will prove the proposition by showing that 𝐹 is bijective on stalks (see
Problem 5-2). It suffices to show that 𝐹𝑈 is bijective whenever 𝑈 ⊆ 𝑀 is an open
subset over which both 𝐸 and 𝐸 ′ are trivial.
Suppose 𝑈 is such a set, and (𝑠𝑗 ), (𝑠′𝑘 ) are holomorphic local frames on 𝑈
for 𝐸 and 𝐸 ′ , respectively. Suppose first that 𝜎 = ∑𝑗 𝜎𝑗 ⊗ 𝜎𝑗′ is an element of
𝒪(𝑈 ; 𝐸) ⊗𝒪(𝑈 ) 𝒪(𝑈 ; 𝐸 ′ ) such that 𝐹𝑈 (𝜎) = 0. Writing 𝜎𝑗 = ∑𝑘 𝑓𝑗𝑘 𝑠𝑘 and 𝜎𝑗′ =
∑𝑙 𝑓𝑗′𝑙 𝑠′𝑙 for some holomorphic functions 𝑓𝑗𝑘 , 𝑓𝑗′𝑙 and using the properties of the
tensor product, we see that the following holds for all 𝑥 ∈ 𝑈 :

0= 𝜎𝑗 (𝑥) ⊗ 𝜎𝑗′ (𝑥) = 𝑓𝑗𝑘 (𝑥)𝑓𝑗′𝑙 (𝑥)𝑠𝑘 (𝑥) ⊗ 𝑠′𝑙 (𝑥).

∑ ∑
𝑗 𝑗,𝑘,𝑙

Since this is true for all 𝑥 ∈ 𝑈 and the elements 𝑠𝑘 (𝑥) ⊗ 𝑠′𝑙 (𝑥) are linearly indepen-
dent, it follows that ∑𝑗 𝑓𝑗𝑘 𝑓𝑗′𝑙 ≡ 0 on 𝑈 for each 𝑘 and 𝑙. Thus

𝜎𝑗 ⊗ 𝜎𝑗′ = 𝑠𝑘 ⊗ 𝑓𝑗𝑘 𝑓𝑗′𝑙 )𝑠′𝑙 = 0,

∑ ∑ ∑(∑
𝑗 𝑘 𝑙 𝑗

so 𝐹𝑈 is injective.
The Étalé Space of a Presheaf 135

On the other hand, if 𝛾 = ∑𝑘,𝑙 𝑔 𝑘𝑙 𝑠𝑘 ⊗ 𝑠′𝑙 is an arbitrary holomorphic section

of 𝐸 ⊗ 𝐸 ′ over 𝑈 , then

𝛾 = 𝐹𝑈 ( 𝑠𝑘 ⊗ ( 𝑔 𝑘𝑙 𝑠′𝑙 )),
∑ ∑
𝑘 𝑙

so 𝐹𝑈 is surjective as well. □

► Exercise 5.13. Given a sheaf ℛ of commutative rings, show that ℛ acts as

an identity under tensor product, in the sense that if 𝒮 is any sheaf of ℛ-modules,
then ℛ ⊗ℛ 𝒮 ≅ 𝒮 ⊗ℛ ℛ ≅ 𝒮 .

Suppose ℛ is a sheaf of commutative rings on a topological space 𝑀. A sheaf

𝒮 of ℛ-modules is called a free sheaf of rank 𝒌 if it is ℛ-module isomorphic to
the 𝑘-fold direct sum ℛ 𝑘 = ℛ ⊕ ⋯ ⊕ ℛ. It is called a locally free sheaf of rank
𝒌 if every point of 𝑀 has a neighborhood 𝑈 on which the restricted sheaf 𝒮 |𝑈 is
isomorphic to ℛ 𝑘 |𝑈 .
The next proposition shows that locally free sheaves of 𝒪-modules of fixed
finite rank on a complex manifold correspond to holomorphic vector bundles.
Proposition 5.14 (Locally Free Sheaves and Vector Bundles). Let 𝑀 be a com-
plex manifold. If 𝐸 → 𝑀 is a holomorphic vector bundle of rank 𝑘, then 𝒪(𝐸)
is a locally free sheaf of 𝒪-modules of rank 𝑘. Conversely, if 𝒮 is a locally free
sheaf of 𝒪-modules of rank 𝑘 on 𝑀, then there is a rank-𝑘 holomorphic vector
bundle 𝐸 → 𝑀, unique up to isomorphism, such that 𝒮 ≅ 𝒪(𝐸) (as sheaves of
𝒪-modules).

Proof. First let 𝐸 → 𝑀 be a holomorphic vector bundle of rank 𝑘. On any open

subset 𝑈 ⊆ 𝑀 over which 𝐸 is trivial, we can choose a holomorphic local frame
(𝑠1 , … , 𝑠𝑘 ). Define a morphism Λ ∶ 𝒪 𝑘 |𝑈 → 𝒪(𝐸)|𝑈 by
Λ𝑉 (𝑓1 , … , 𝑓𝑘 ) = 𝑓1 𝑠1 + ⋯ + 𝑓𝑘 𝑠𝑘 ∈ 𝒪(𝑉 ; 𝐸)
for each open subset 𝑉 ⊆ 𝑈 and holomorphic functions 𝑓1 , … , 𝑓𝑘 ∈ 𝒪(𝑉 ). This
is easily verified to be an 𝒪-module isomorphism.
Conversely, suppose 𝒮 is a locally free sheaf of 𝒪-modules of rank 𝑘 on 𝑀.
There is an open cover {𝑈𝛼 }𝛼∈𝐴 of 𝑀 such that for each 𝛼 ∈ 𝐴 there exists an
isomorphism Ψ𝛼 ∶ 𝒮 |𝑈𝛼 ≅ 𝒪 𝑘 |𝑈𝛼 . Where two such open sets 𝑈𝛼 , 𝑈𝛽 intersect, we
have two isomorphisms of 𝒪(𝑈𝛼 ∩ 𝑈𝛽 )-modules:
(Ψ𝛽 )𝑈𝛼 ∩𝑈𝛽 (Ψ𝛼 )𝑈𝛼 ∩𝑈𝛽
𝒪(𝑈𝛼 ∩ 𝑈𝛽 )𝑘 ←−−−−−−− 𝒮 (𝑈𝛼 ∩ 𝑈𝛽 ) −−−−−−−→ 𝒪(𝑈𝛼 ∩ 𝑈𝛽 )𝑘 .
Because the composite map (Ψ𝛼 )𝑈𝛼 ∩𝑈𝛽 ∘ (Ψ𝛽 )−1
𝑈𝛼 ∩𝑈𝛽 is an 𝒪(𝑈𝛼 ∩ 𝑈𝛽 )-module iso-
morphism, it has the form
(𝑓1 , … , 𝑓𝑘 ) ↦ (𝑡1 𝑓𝑗 , … , 𝑡𝑘 𝑓𝑗 )
𝑗 𝑗
(5.9)
136 5. Sheaves

for some invertible 𝑘 × 𝑘 matrix of functions 𝑡𝑖 ∈ 𝒪(𝑈𝛼 ∩ 𝑈𝛽 ). Let us denote this

𝑗

matrix-valued holomorphic function by 𝜏𝛼𝛽 . A straightforward computation shows

that 𝜏𝛼𝛽 𝜏𝛽𝛾 = 𝜏𝛼𝛾 on 𝑈𝛼 ∩ 𝑈𝛽 ∩ 𝑈𝛾 whenever the intersection is nonempty. Thus by
the vector bundle construction theorem, the collection of functions {𝜏𝛼𝛽 ∶ 𝛼, 𝛽 ∈
𝐴} forms the data for a holomorphic vector bundle 𝐸 → 𝑀 that is trivial over each
set 𝑈𝛼 .
To see that the sheaf 𝒪(𝐸) is isomorphic to 𝒮 , note that each local trivialization
Φ𝛼 ∶ 𝐸|𝑈𝛼 → 𝑈𝛼 × ℂ𝑘 induces a sheaf isomorphism Φ ̃ 𝛼 ∶ 𝒪(𝐸)|𝑈 → 𝒪 𝑘 |𝑈 (the
𝛼 𝛼

𝑈𝛼 and 𝑈𝛽 intersect, these isomorphisms satisfy

inverse of the isomorphism constructed in the first paragraph of this proof). Where

̃ 𝛼 )𝑈 ∩𝑈 ∘ (Φ
(Φ ̃ 𝛽 )−1 (𝑓1 , … , 𝑓𝑘 ) = ((𝜏𝛼𝛽 ) 𝑓𝑗 , … , (𝜏𝛼𝛽 ) 𝑓𝑗 ).
𝑗 𝑗
(5.10) 𝛼 𝛽 𝑈 ∩𝑈
𝛼 𝛽 1 𝑘

Thus restricted to 𝑈𝛼 ∩ 𝑈𝛽 , it follows from (5.9) and (5.10) that

(Ψ𝛼 )𝑈𝛼 ∩𝑈𝛽 ∘ (Ψ𝛽 )−1 ̃ ̃ −1

(5.11) 𝑈 ∩𝑈 = (Φ𝛼 )𝑈𝛼 ∩𝑈𝛽 ∘ (Φ𝛽 )𝑈 ∩𝑈 .
𝛼 𝛽 𝛼 𝛽

For each 𝛼, define a sheaf isomorphism 𝐹𝛼 ∶ 𝒪(𝐸)|𝑈𝛼 → 𝒮 |𝑈𝛼 by

𝐹𝛼 = Ψ−1 ̃
𝛼 ∘ Φ𝛼 .

It follows from (5.11) that the restrictions of 𝐹𝛼 and 𝐹𝛽 agree on 𝑈𝛼 ∩ 𝑈𝛽 . Thus

we can define a sheaf isomorphism 𝐹 ∶ 𝒪(𝐸) → 𝒮 as follows: for each open set
𝑉 ⊆ 𝑀 and section 𝜎 ∈ 𝒪(𝑉 ; 𝐸), let 𝐹𝑉 (𝜎) ∈ 𝒮 (𝑉 ) be the section satisfying

𝐹𝑉 (𝜎)|𝑉 ∩𝑈 = (𝐹𝛼 )𝑉 ∩𝑈𝛼 (𝜎 |𝑉 ∩𝑈 ) for each 𝛼 ∈ 𝐴.

𝛼 𝛼

The gluing and locality properties of 𝒮 guarantee that this is well defined.
To show that the bundle 𝐸 is unique up to isomorphism, it suffices to show
that if 𝐸 and 𝐸 ′ are rank-𝑘 holomorphic vector bundles over 𝑀 such that 𝒪(𝐸) ≅
𝒪(𝐸 ′ ), then 𝐸 ≅ 𝐸 ′ . Given such bundles 𝐸 and 𝐸 ′ , suppose 𝐹 ∶ 𝒪(𝐸) → 𝒪(𝐸 ′ ) is
a sheaf isomorphism. We can choose an open cover {𝑈𝛼 }𝛼∈𝐴 of 𝑀 such that 𝐸 and
𝐸 ′ are both trivial over each 𝑈𝛼 , with local trivializations Φ𝛼 and Φ′𝛼 and transition
functions 𝜏𝛼𝛽 and 𝜏𝛼𝛽 ′

̃
Φ𝛼 ∶ 𝒪(𝐸)|𝑈𝛼 → 𝒪 |𝑈𝛼 and Φ ̃ ′𝛼 ∶ 𝒪(𝐸 ′ )|𝑈 → 𝒪 𝑘 |𝑈 . For each 𝛼 ∈ 𝐴, we have a
. As above, these isomorphisms lead to sheaf isomorphisms
𝑘

composition of 𝒪(𝑈𝛼 )-module isomorphisms

𝛼 𝛼

̃ 𝛼 )−1
(Φ (𝐹 )𝑈𝛼 ̃ ′𝛼 )𝑈
(Φ
𝒪(𝑈𝛼 )𝑘 −−−−−→ 𝒪(𝑈𝛼 ; 𝐸) −−−−→ 𝒪(𝑈𝛼 ; 𝐸 ′ ) −−−−−→ 𝒪(𝑈𝛼 )𝑘 .
𝑈
𝛼 𝛼
Exact Sequences of Sheaves 137

function 𝜓𝛼 ∶ 𝑈𝛼 → GL(𝑘, ℂ), and a straightforward computation shows that these

As before, each such composition is represented by a matrix-valued holomorphic

functions satisfy 𝜏𝛼𝛽 = 𝜓𝛼−1 𝜏𝛼𝛽

′
𝜓𝛽 on 𝑈𝛼 ∩ 𝑈𝛽 , so Proposition 3.7 shows that the
corresponding bundles are isomorphic. □

► Exercise 5.15. Verify that the same proof shows there are analogous corre-
spondences between locally free sheaves of ℰ -modules and smooth vector bun-
dles on a smooth manifold, and between locally free sheaves of 𝒞 -modules and
topological vector bundles on a topological space.

A locally free sheaf (of 𝒪-modules, say) of rank 1 is called an invertible sheaf ,
reflecting the fact that every such sheaf has an inverse under tensor product: if 𝐿 is a

that 𝒪(𝐿) ⊗𝒪 𝒪(𝐿∗ ) ≅ 𝒪(𝐿 ⊗ 𝐿∗ ) ≅ 𝒪.

holomorphic line bundle, then it follows from Proposition 5.12 and Lemma 3.28(d)

free sheaves of 𝒪-modules, with vector bundles relegated to the background. If

In the algebraic geometry literature, it is common to focus primarily on locally

sponding sheaves. In particular, on ℂℙ𝑛 , the sheaf 𝒪(𝐻 𝑑 ) of holomorphic sections

vector bundles are mentioned at all, they are often not distinguished from the corre-

of the 𝑑-fold tensor power of the hyperplane bundle is usually abbreviated by 𝒪(𝑑),

line bundle 𝐻 𝑑 itself. From a differential-geometric point of view, however, vector

and that shorthand notation is used both for the invertible sheaf of sections and the

bundles and sheaves are very different objects, and we will continue to use different
notations for them.
The next proposition expresses an extremely important relationship between
ideal sheaves and invertible sheaves.
Proposition 5.16. Suppose 𝑀 is a complex manifold, 𝑆 ⊆ 𝑀 is a closed complex
hypersurface, 𝐿𝑆 → 𝑀 is the holomorphic line bundle associated with 𝑆, and
𝐸 → 𝑀 is any holomorphic vector bundle.
(a) The ideal sheaf ℐ𝑆 is isomorphic to the sheaf 𝒪(𝐿∗𝑆 ) of holomorphic sec-
tions of the dual bundle 𝐿∗𝑆 .
(b) More generally, the sheaf ℐ𝑆 (𝐸) of holomorphic sections of 𝐸 vanishing
on 𝑆 is isomorphic to ℐ𝑆 ⊗𝒪 𝒪(𝐸) ≅ 𝒪(𝐿∗𝑆 ⊗ 𝐸).
(c) The sheaf ℐ𝑆2 (𝐸) of holomorphic sections of 𝐸 that vanish to second order
on 𝑆 is isomorphic to 𝒪(𝐿∗𝑆 ⊗ 𝐿∗𝑆 ⊗ 𝐸).

Proof. Problem 5-10. □

Exact Sequences of Sheaves

Most of the applications of sheaves are based on the notion of exact sequences.
For abelian groups, rings, or vector spaces, the definition of an exact sequence is
138 5. Sheaves

straightforward—it is a sequence of homomorphisms such that the image of each

homomorphism is equal to the kernel of the next. But that simple characterization

If 𝐹 ∶ 𝒮 → 𝒯 is a morphism between sheaves of abelian groups, we can define

fails to make sense for sheaves.

a subsheaf Ker 𝐹 ⊆ 𝒮 , called the kernel of 𝑭 , by

(Ker 𝐹 )(𝑈 ) = Ker 𝐹𝑈 ⊆ 𝒮 (𝑈 ).

► Exercise 5.17. In the situation described above, show that Ker 𝐹 satisfies the
gluing and locality properties, and thus is a subsheaf of 𝒮 .

presheaf 𝑈 ↦ Im 𝐹𝑈 may fail to be a sheaf, as the following example demonstrates.

However, for the image of a sheaf morphism, things are not so easy, because the

Example 5.18. Let 𝑀 be a smooth 𝑛-manifold, and for each nonnegative integer
𝑘, let ℰ 𝑘 be the sheaf of smooth 𝑘-forms on 𝑀. Let 𝑑 ∶ ℰ 𝑘 → ℰ 𝑘+1 be the sheaf
morphism defined by exterior differentiation. The image presheaf of 𝑑 is the pre-
sheaf for which (Im 𝑑)(𝑈 ) is the space of exact (𝑘 + 1)-forms on 𝑈 . However, the
gluing property might fail for this presheaf: a form that is locally exact need not be
globally exact. //

of sheaf morphisms. Consider sheaves ℛ, 𝒮 , 𝒯 of abelian groups on a topological

To remedy this problem, we need a special definition of exactness for sequences

space 𝑀 together with sheaf morphisms

𝐹 𝐺
(5.12) ℛ ⟶𝒮 ⟶𝒯.
The sequence is said to be exact if it is exact on each stalk: for each 𝑝 ∈ 𝑀, the
sequence of induced stalk homomorphisms
𝐹𝑝 𝐺𝑝
(5.13) ℛ𝑝 ⟶ 𝒮 𝑝 ⟶ 𝒯 𝑝
is exact in the usual sense that the image of 𝐹𝑝 is equal to the kernel of 𝐺𝑝 .
The next exercise shows that exactness on the presheaf level implies exactness
on the sheaf level.

► Exercise 5.19. Suppose ℛ, 𝒮 , and 𝒯 are presheaves of abelian groups on a

topological space 𝑀, and 𝐹 ∶ ℛ → 𝒮 and 𝐺 ∶ 𝒮 → 𝒯 are morphisms such that
the following sequence is exact for each open subset 𝑈 ⊆ 𝑀:
𝐹𝑈 𝐺𝑈
ℛ(𝑈 ) ⟶ 𝒮 (𝑈 ) ⟶ 𝒯 (𝑈 ).
Prove that the associated sheaf sequence

ℛ+ ⟶ 𝒮 + ⟶ 𝒯 +
𝐹+ 𝐺+

is exact.
Exact Sequences of Sheaves 139

But sheaf sequences that are not exact on the presheaf level may still be exact
as sheaf sequences. It is worth unpacking the definition to see what this means
explicitly in practice.

Lemma 5.20. The sheaf sequence (5.12) is exact if and only if both of the following
conditions are satisfied:

(i) 𝐺𝑈 ∘ 𝐹𝑈 = 0 for every open set 𝑈 ⊆ 𝑀, and

(ii) Given 𝑠 ∈ 𝒮 (𝑈 ) such that 𝐺𝑈 (𝑠) = 0, for each 𝑝 ∈ 𝑈 there are an open
set 𝑉 ⊆ 𝑈 containing 𝑝 and a section 𝑟 ∈ ℛ(𝑉 ) such that 𝐹𝑉 (𝑟) = 𝑠|𝑉 .

the stalk homomorphism 𝐹𝑝 is contained in Ker 𝐺𝑝 for each 𝑝. To prove the reverse
Proof. First assume that (i) and (ii) hold. Condition (i) implies that the image of

inclusion, let [𝑠]𝑝 ∈ Ker 𝐺𝑝 be arbitrary. Then there is some neighborhood 𝑈 of 𝑝

and a section 𝑠 representing this germ such that 𝐺𝑈 (𝑠) = 0 ∈ 𝒯 (𝑈 ). Condition (ii)
guarantees the existence of a smaller neighborhood 𝑉 of 𝑝 such that 𝑠|𝑉 = 𝐹𝑉 (𝑟)
for some section 𝑟 ∈ ℛ(𝑉 ). It follows that 𝐹𝑝 ([𝑟]𝑝 ) = [𝐹𝑉 (𝑟)]𝑝 = [𝑠]𝑝 , so Ker 𝐺𝑝 ⊆
Im 𝐹𝑝 .
Conversely, assume the sequence (5.12) is exact. Suppose 𝑟 ∈ ℛ(𝑈 ) for some
open set 𝑈 . Then for each 𝑝 ∈ 𝑈 , the germ of 𝐺𝑈 ∘ 𝐹𝑈 (𝑟) at 𝑝 represents an
element of 𝐺𝑝 ∘ 𝐹𝑝 (ℛ𝑝 ), which is zero. This means 𝐺𝑈 ∘ 𝐹𝑈 (𝑟) restricts to zero in
a neighborhood of each point, so by the locality property it is zero. This shows that
(i) holds. The proof of (ii) is left as an exercise. □

► Exercise 5.21. Complete the proof of this lemma by showing that if the se-
quence (5.12) is exact, then condition (ii) holds.

An important special case is a short exact sequence of sheaves: this is a five-

term exact sequence of the form

𝐹 𝐺
(5.14) 0 → ℛ ⟶ 𝒮 ⟶ 𝒯 → 0,

where the zeros on the ends represent the trivial sheaf , whose spaces of sections

Lemma 5.20 shows that exactness at ℛ means that whenever 𝑟 ∈ ℛ(𝑈 ) is in

and restriction maps are all zero.

the kernel of 𝐹𝑈 , it restricts to zero in a neighborhood of each point; but in this

case, since there is only one zero section, it follows that 𝐹𝑈 ∶ ℛ(𝑈 ) → 𝒮 (𝑈 ) is
injective for each 𝑈 . If this is the case, we say the sheaf morphism 𝐹 is injective.
On the other hand, exactness at 𝒯 means that given 𝑡 ∈ 𝒯 (𝑈 ), for each 𝑝 ∈ 𝑈
there is a neighborhood 𝑉 of 𝑝 contained in 𝑈 and a section 𝑠 ∈ 𝒮 (𝑉 ) such that
𝐺𝑉 (𝑠) = 𝑡|𝑉 ; in this case, we say the sheaf morphism 𝐺 is surjective.
140 5. Sheaves

Proposition 5.22. Suppose 𝑀 is a topological space.

(a) If 𝒮 is a sheaf of abelian groups on 𝑀 and ℛ is a subsheaf of 𝒮 , there is
a canonical sheaf morphism Π ∶ 𝒮 → 𝒮 /ℛ such that the following sheaf
sequence is exact:
𝑖 Π
(5.15) 0 → ℛ ↪ 𝒮 ⟶ 𝒮 /ℛ → 0,

where 𝑖 is the inclusion of ℛ into 𝒮 .

of abelian groups on 𝑀, 𝐹 is an isomorphism onto Ker 𝐺, and there is a

(b) For any short exact sequence of sheaf morphisms (5.14) between sheaves

sheaf isomorphism 𝐺 ∶ 𝒮 / Ker 𝐺 → 𝒯 such that the following diagram

commutes:

0 /ℛ 𝐹 /𝒮 𝒯O / 0.
𝐺 /

𝐺
❉❉

Π ❉❉!
(5.16)
❉❉

𝒮 / Ker 𝐺
❉❉

Proof. Given a subsheaf ℛ ↪ 𝒮 , let 𝒫 be the presheaf 𝒫 (𝑈 ) = 𝒮 (𝑈 )/ℛ(𝑈 ),

whose sheafification is the quotient sheaf 𝒮 /ℛ; and let 𝜋 ∶ 𝒮 → 𝒫 be the presheaf
morphism given by the quotient maps 𝜋𝑈 ∶ 𝒮 (𝑈 ) → 𝒮 (𝑈 )/ℛ(𝑈 ). We define the
sheaf morphism Π ∶ 𝒮 → 𝒮 /ℛ by Π = 𝜃𝒫 ∘ 𝜋, where 𝜃𝒫 ∶ 𝒫 → 𝒮 /ℛ is the

The sequence (5.15) is exact at ℛ because 𝑖 is injective. For exactness at 𝒮 ,

canonical presheaf morphism given by Theorem 5.9.

we will use Lemma 5.20. Note first that 𝜋 ∘ 𝑖 = 0, which implies Π ∘ 𝑖 = 0, so

condition (i) of the lemma is satisfied. On the other hand, given 𝑠 ∈ 𝒮 (𝑈 ) such
that Π𝑈 (𝑠) = 0, this means that for each 𝑝 ∈ 𝑈 there is a neighborhood 𝑉 of 𝑝 such
that 𝑠|𝑉 ∈ ℛ(𝑉 ), so condition (ii) is satisfied.
Finally, exactness at 𝒮 /ℛ is just the fact that each element 𝜉 of the stalk (𝒮 /ℛ)𝑝
is represented on some neighborhood 𝑉 of 𝑝 by some section 𝑠 ∈ 𝒮 (𝑉 ) mod ℛ(𝑉 ),
and it follows that Π𝑝 ([𝑠]𝑝 ) = 𝜉. This completes the proof of (a).
To prove the first claim in (b), let 𝒦 denote the subsheaf Ker 𝐺 ⊆ 𝒮 . For each
open set 𝑈 ⊆ 𝑀, Lemma 5.20 shows that 𝐹𝑈 (ℛ(𝑈 )) ⊆ 𝒦 (𝑈 ), and the discus-
sion in the paragraph preceding the statement of the proposition showed that 𝐹𝑈 is
injective. On the other hand, given 𝑘 ∈ 𝒦 (𝑈 ), each 𝑝 ∈ 𝑈 has a neighborhood
𝑉 ⊆ 𝑈 on which 𝑘|𝑉 = 𝐹𝑉 (𝑟) for some 𝑟 ∈ ℛ(𝑉 ). The collection of all such
neighborhoods is an open cover of 𝑈 , and wherever two such sets 𝑉 , 𝑉 ′ overlap
and 𝑟, 𝑟′ are the corresponding sections, we have

𝐹𝑉 ∩𝑉 ′ (𝑟|𝑉 ∩𝑉 ′ ) = 𝐹𝑉 (𝑟)|𝑉 ∩𝑉 ′ = 𝑘|𝑉 ∩𝑉 ′ = 𝐹𝑉 ′ (𝑟′ )|𝑉 ∩𝑉 ′

= 𝐹𝑉 ∩𝑉 ′ (𝑟′ |𝑉 ∩𝑉 ′ ),
Exact Sequences of Sheaves 141

and injectivity of 𝐹𝑉 ∩𝑉 ′ shows that 𝑟|𝑉 ∩𝑉 ′ = 𝑟′ |𝑉 ∩𝑉 ′ . Thus these sections glue

together to form a section 𝑟 ̃ ∈ ℛ(𝑈 ) satisfying 𝐹𝑈 (𝑟)̃ = 𝑘, showing that the homo-
morphism 𝐹𝑈 ∶ ℛ(𝑈 ) → 𝒦 (𝑈 ) is also surjective.
To prove the second claim in (b), note first that the existence of the morphism 𝐺
making (5.16) commute follows from the result of Problem 5-8(b) (identifying 𝒯
with the quotient sheaf 𝒯 /0). We will show that for each open subset 𝑈 ⊆ 𝑀, the
group homomorphism 𝐺𝑈 ∶ (𝒮 /𝒦 )(𝑈 ) → 𝒯 (𝑈 ) is bijective. From the definition
of the sheafification functor, for any sections 𝜎 ∈ (𝒮 /𝒦 )(𝑈 ) and 𝜏 ∈ 𝒯 (𝑈 ), the
equation 𝐺𝑈 (𝜎) = 𝜏 means that for each 𝑝 ∈ 𝑈 there exist a neighborhood 𝑉𝑝
and a section 𝑠𝑝 ∈ 𝒮 (𝑉𝑝 )/𝒦 (𝑉𝑝 ) such that 𝐺𝑉𝑝 (𝑠𝑝 ) = 𝜏|𝑉𝑝 . To see that 𝐺𝑈 is
injective, if 𝐺𝑈 (𝜎) = 0, then because each 𝐺𝑉𝑝 is injective on the quotient group
𝒮 (𝑉𝑝 )/𝒦 (𝑉𝑝 ), we see that 𝜎 restricts to zero on each 𝑉𝑝 and thus is zero itself. To
see that 𝐺𝑈 is surjective, given 𝜏 ∈ 𝒯 (𝑈 ), the surjectivity of 𝐺 means that for each
𝑝 ∈ 𝑈 we can choose a neighborhood 𝑉𝑝 small enough that there exists 𝑠𝑝 ∈ 𝒮 (𝑉𝑝 )
with 𝐺𝑉𝑝 (𝑠𝑝 ) = 𝜏|𝑉𝑝 , and any two such sections agree modulo 𝒦 (𝑉𝑝 ); thus the
local sections can be glued together to form a section 𝜎 ∈ (𝒮 /𝒦 )(𝑈 ) satisfying
𝐺𝑈 (𝜎) = 𝜏. □

Example 5.23 (Exact Sheaf Sequences).

(a) On an 𝑛-dimensional smooth manifold 𝑀, we have a sequence of sheaves
of complex vector spaces
𝑑 𝑑 𝑑 𝑑
(5.17) 0 → ℂ ↪ ℰ 0 ⟶ ℰ 1 ⟶ ℰ 2 ⟶ ⋯ ⟶ ℰ 𝑛 → 0.

image of 𝑑.
It is an exact sheaf sequence because every closed form is locally in the

(b) For each 𝑘 ≥ 0, we can extract a short exact sheaf sequence from the one
above:
𝑑
(5.18) 0 → 𝒵 𝑘 ↪ ℰ 𝑘 ⟶ 𝒵 𝑘+1 → 0,

where 𝒵 𝑘 and 𝒵 𝑘+1 are the sheaves of closed forms.

(c) On a complex 𝑛-manifold 𝑀, thanks to the 𝜕-Poincaré lemma, for each 𝑝
we have an exact sheaf sequence associated with the Dolbeault complex:
𝜕 𝜕
(5.19) 0 → Ω𝑝 ↪ ℰ 𝑝,0 ⟶ ℰ 𝑝,1 ⟶ ℰ 𝑝,2 → ⋯ → ℰ 𝑝,𝑛 → 0,

where Ω𝑝 is the sheaf of holomorphic 𝑝-forms (i.e., 𝜕-closed (𝑝, 0)-forms).

As before, we can extract short exact sequences:
𝜕
0 → 𝒵 𝑝,𝑞 ↪ ℰ 𝑝,𝑞 ⟶ 𝒵 𝑝,𝑞+1 → 0.
142 5. Sheaves

More generally, if 𝐸 → 𝑀 is a holomorphic vector bundle, then we have

an exact sheaf sequence of bundle-valued (𝑝, 𝑞)-forms:
𝜕𝐸 𝜕𝐸
0 → Ω𝑝 (𝐸) ↪ ℰ 𝑝,0 (𝐸) ⟶ ℰ 𝑝,1 (𝐸) ⟶ ℰ 𝑝,2 (𝐸) → ⋯ → ℰ 𝑝,𝑛 (𝐸) → 0.
(d) On a complex manifold 𝑀, we have the following sequence of sheaves of
abelian groups:
𝜄 𝜀
(5.20) 0 → ℤ ↪ 𝒪 ⟶ 𝒪 ∗ → 0,
where ℤ is the constant sheaf, 𝜄 is inclusion of integer-valued functions as
holomorphic functions, and 𝜀(𝑓 ) = 𝑒2𝜋𝑖𝑓 . This is called the exponential
sheaf sequence. It is immediate that 𝜄 is injective and 𝜀 ∘ 𝜄(𝑓 ) ≡ 1. (By
custom, we denote the right-hand trivial sheaf by 0, even though 𝒪 ∗ is a

urally written as 1.) To check exactness at 𝒪, just note that if 𝑒2𝜋𝑖𝑓 = 1

sheaf of groups under multiplication whose identity section is more nat-

on an open set 𝑈 , then 𝑓 must be constant and integer-valued on each

connected component of 𝑈 . Surjectivity at 𝒪 ∗ follows from the fact that
each point of 𝑀 has a neighborhood on which every nonvanishing holo-
morphic function has a holomorphic logarithm (see Problem 5-11).
(e) On a smooth manifold, there is a sheaf sequence analogous to the one
above, the smooth exponential sheaf sequence:
𝜄 𝜀
0 → ℤ ↪ ℰ ⟶ ℰ ∗ → 0,
where ℰ and ℰ ∗ are the sheaves of smooth complex-valued functions and
nonvanishing smooth complex-valued functions, respectively. It is exact
by the same argument.
(f) Suppose 𝑀 is a complex manifold and 𝑝 ∈ 𝑀. We have a short exact
sheaf sequence
𝑒
(5.21) 0 → ℐ{𝑝} ↪ 𝒪 ⟶ ℂ𝑝 → 0,
where ℐ{𝑝} is the ideal sheaf of {𝑝} (the sheaf of holomorphic functions
that vanish at 𝑝), ℂ𝑝 is the skyscraper sheaf whose stalk at 𝑝 is ℂ and all
other stalks are zero, and 𝑒 is the evaluation map 𝑒𝑈 (𝑓 ) = 𝑓 (𝑝) if 𝑝 ∈ 𝑈
and 0 otherwise. //
The basic fact about sheaves is that even if a sheaf sequence is exact, the local
nature of sheaf exactness means that exactness does not necessarily hold on spaces
of sections. For example, in the sheaf sequence (5.17), if we look at the spaces of
global sections, we find a sequence of vector spaces and linear maps
𝑑 𝑑
(5.22) 0 → ℰ 0 (𝑀) ⟶ ℰ 1 (𝑀) ⟶ ℰ 2 (𝑀) → ⋯ → ℰ 𝑛 (𝑀) → 0.
Exact Sequences of Sheaves 143

This sequence is not exact, but it is a cochain complex. Its cohomology groups are

𝐻dR𝑘
(𝑀; ℂ). To avoid confusion, we will denote ordinary (real) de Rham cohomol-
the de Rham cohomology groups with complex coefficients, which we denote by

ogy by 𝐻dR𝑘
(𝑀; ℝ). Because conjugation of complex differential forms commutes
with the exterior derivative operator, conjugation descends to a conjugate-linear

𝐻dR𝑘
(𝑀; ℝ) as a real-linear subspace of 𝐻dR 𝑘
(𝑀; ℂ), namely the space of coho-
automorphism of each complex de Rham cohomology group. Thus we can view

mology classes that are invariant under conjugation, and 𝐻dR 𝑘

(𝑀; ℂ) is isomorphic
to the complexification of 𝐻dR (𝑀; ℝ). It follows that the complex dimension of
𝑘

𝐻dR𝑘
(𝑀; ℂ) is equal to the real dimension of 𝐻dR 𝑘
(𝑀; ℝ), namely the 𝑘th Betti
number 𝑏 (𝑀).
𝑘

We can also look at the global section sequence associated with the short exact
sequence (5.18):

𝑑
0 → 𝒵 𝑘 (𝑀) ↪ ℰ 𝑘 (𝑀) ⟶ 𝒵 𝑘+1 (𝑀).

This global section sequence is exact as far as it goes, as you can check (or see Prop.

have left off the last zero. In fact, the question of when 𝑑 is surjective onto the space
5.24 below), but the last homomorphism need not be surjective, which is why we

of closed forms, and if not, how to quantify the failure of exactness, is exactly the

For another example, on a complex manifold 𝑀, let 𝑝, 𝑞 be distinct points of

question addressed by de Rham cohomology.

𝑀 and consider the following variation on the sequence (5.21):

𝑒
0 → ℐ{𝑝,𝑞} ↪ 𝒪 ⟶ ℂ𝑝 ⊕ ℂ𝑞 → 0,

where now 𝑒 evaluates a function at the two points 𝑝 and 𝑞. (The sheaf ℂ𝑝 ⊕ ℂ𝑞 is
a “double skyscraper sheaf”: it has two nontrivial stalks and all the rest are zero.)
When we examine the global section maps, we find

𝑒𝑀
0 → ℐ{𝑝,𝑞} (𝑀) ↪ 𝒪(𝑀) ⟶ ℂ ⊕ ℂ,

where 𝑒𝑀 (𝑓 ) = (𝑓 (𝑝), 𝑓 (𝑞)). Again, you can check that this sequence is exact as
far as it goes, but the question of whether 𝑒𝑀 is surjective for all 𝑝 and 𝑞 is exactly
the question of whether 𝒪(𝑀) separates points, a key requirement for 𝑀 to be a
Stein manifold. A variant of this sheaf sequence will play a central role in our proof
of the Kodaira embedding theorem in Chapter 10.
In the last two examples, global section sequences associated with short exact
sheaf sequences failed to be exact, but it was only surjectivity at the last term that
failed. The next proposition shows that this pattern is quite general.
144 5. Sheaves

Proposition 5.24. On a topological space 𝑀, suppose the following is an exact

sequence of sheaves of abelian groups
𝐹 𝐺
(5.23) 0 → ℛ ⟶ 𝒮 ⟶ 𝒯 → 0.
Then for each open subset 𝑈 ⊆ 𝑀, the section sequence
𝐹𝑈 𝐺𝑈
(5.24) 0 → ℛ(𝑈 ) ⟶ 𝒮 (𝑈 ) ⟶ 𝒯 (𝑈 )
is exact.

Proof. As we noted above, exactness of (5.23) at ℛ implies that 𝐹𝑈 ∶ ℛ(𝑈 ) →

𝒮 (𝑈 ) is injective for all 𝑈 , so (5.24) is exact at ℛ(𝑈 ). To prove exactness at 𝒮 (𝑈 ),
note that Proposition 5.22 showed that 𝐹 is an isomorphism from ℛ onto the sub-
sheaf Ker 𝐺 ⊆ 𝒮 ; in particular, this means that 𝐹𝑈 (ℛ(𝑈 )) = Ker(𝐺𝑈 ). □

In the next chapter, we will develop some powerful tools to help determine
when the global section sequence associated with a short exact sheaf sequence is
exact.
Most of the sheaves that arise in complex geometry are of three types: constant

sheaves of 𝒪-modules (called analytic sheaves), which carry information about the
sheaves, which carry topological information (see, for example, Thm. 6.18 below);

holomorphic structure (e.g., Thm. 6.19); and sheaves of ℰ -modules, which provide
a crucial tool for deriving properties of the other two types (e.g., Thm. 6.11). The
analytic sheaves that we will be able to say the most about are the locally free ones,
that is, the sheaves of sections of holomorphic vector bundles.
It is worth remarking that there is a generalization of locally free sheaves that
turns out to be extremely important in algebraic geometry and analysis of several

as locally free sheaves. An analytic sheaf 𝒮 on a complex manifold 𝑀 is said to be

complex variables, because sheaves of this type have many of the same properties

coherent if it is locally finitely generated, meaning each point of 𝑀 has a neigh-

borhood 𝑈 and finitely many sections 𝑠1 , … , 𝑠𝑚 ∈ 𝒮 (𝑈 ) that generate each stalk
𝒮𝑥 as an 𝒪𝑥 -module for 𝑥 ∈ 𝑈 ; and given any such sections, the kernel of the sheaf
morphism 𝒪 𝑚 |𝑈 → 𝒮 |𝑈 given by (𝑓1 , … , 𝑓𝑚 ) ↦ ∑𝑖 𝑓𝑖 𝑠𝑖 is also locally finitely
generated (where, as before, 𝒪 𝑚 = 𝒪 ⊕ ⋯ ⊕ 𝒪). Locally free analytic sheaves are
coherent, as are ideal sheaves associated with analytic or algebraic varieties (see
[GH94, pp. 695–704]). We will not have occasion to use coherent analytic sheaves
in this book.

Problems
5-1. Suppose 𝒮 is a sheaf of abelian groups and ℛ ↪ 𝒮 is a subsheaf. Show
that each stalk homomorphism ℛ𝑝 → 𝒮𝑝 is injective, so it makes sense
to identify ℛ𝑝 with a subgroup of 𝒮𝑝 .
Problems 145

5-2. Suppose 𝐹 ∶ 𝒮 → 𝒯 is a morphism between sheaves of abelian groups

over a topological space 𝑀. Show that 𝐹 is an isomorphism if and only
if each stalk homomorphism 𝐹𝑝 is bijective.
5-3. Let 𝑀 be a topological space and 𝐺 an abelian group. Define the constant
presheaf on 𝑴 with coefficients in 𝑮 to be the presheaf 𝒢 for which
𝒢 (𝑈 ) = 𝐺 for every open set 𝑈 , and all restriction maps are the identity.
Show that the sheafification of 𝒢 is isomorphic to the constant sheaf 𝐺.
5-4. Suppose 𝒮 → 𝑀 is an étalé space, 𝑈 ⊆ 𝑀 is open, and 𝜎, 𝜏 ∶ 𝑈 → 𝒮
are continuous sections. Show that the set {𝑥 ∈ 𝑈 ∶ 𝜎(𝑥) = 𝜏(𝑥)} is
open. If 𝑈 is connected and 𝜎(𝑝) = 𝜏(𝑝) for some 𝑝 ∈ 𝑀, does this imply
that 𝜎 ≡ 𝜏 on 𝑈 ?
5-5. Prove Proposition 5.10 (the étalé space functor is an equivalence of cate-
gories).
(a) Let 𝒞 be the sheaf of continuous real-valued functions on ℝ. Show
that the étalé space Et(𝒞 ) is not Hausdorff.
5-6.

(b) Let 𝒪 be the sheaf of holomorphic functions on a complex manifold.

Show that Et(𝒪) is Hausdorff.
5-7. Suppose 𝒮 and 𝒯 are sheaves of abelian groups on a topological space
𝑀. Define a presheaf ℋom(𝒮 , 𝒯 ) by letting ℋom(𝒮 , 𝒯 )(𝑈 ) be the group
of sheaf morphisms from 𝒮 |𝑈 to 𝒯 |𝑈 . Show that ℋom(𝒮 , 𝒯 ) is a sheaf
of abelian groups on 𝑀.
5-8. Suppose 𝒮 , 𝒮 ′ are sheaves of abelian groups on a topological space 𝑀
and ℛ ↪ 𝒮 , ℛ ′ ↪ 𝒮 ′ are subsheaves.
(a) Show that each stalk (𝒮 /ℛ)𝑝 is canonically isomorphic to the quo-
tient group 𝒮𝑝 /ℛ𝑝 , where ℛ𝑝 is considered as a subgroup of 𝒮𝑝 as in

(b) Show that if 𝐹 ∶ 𝒮 → 𝒮 ′ is a morphism that takes ℛ to ℛ ′ , then 𝐹

Exercise 5.6.

passes to the quotient to yield a morphism 𝐹 ∶ 𝒮 /ℛ → 𝒮 ′ /ℛ ′ .

5-9. Let 𝑀 be a complex manifold and let 𝐿 → 𝑀 be a holomorphic line

the tautological bundle on ℂℙ𝑛 ). Show that the presheaf 𝒯 defined by

bundle that admits no nontrivial global holomorphic sections (such as

𝒯 (𝑈 ) = 𝒪(𝑈 ; 𝐿) ⊗𝒪(𝑈 ) 𝒪(𝑈 ; 𝐿∗ ) is not a sheaf because it does not sat-

isfy the gluing property.
5-10. Prove Proposition 5.16 (relating ideal sheaves and sheaves of sections of
line bundles).
5-11. Suppose 𝑈 ⊆ ℂ𝑛 is a simply connected open set and 𝑓 ∶ 𝑈 → ℂ is

𝐿 ∶ 𝑈 → ℂ such that 𝑓 (𝑧) = 𝑒𝐿(𝑧) for all 𝑧 ∈ 𝑈 ; and show that 𝐿 is

a nonvanishing smooth function. Prove that there is a smooth function

holomorphic if 𝑓 is. [Hint: Consider the closed 1-form 𝜔 = 𝑑𝑓 /𝑓 .]

146 5. Sheaves

5-12. Let 𝑋 and 𝑌 be topological spaces and 𝑓 ∶ 𝑋 → 𝑌 a continuous map. For

any sheaf 𝒮 on 𝑋 (of abelian groups, say), define a presheaf 𝑓∗ 𝒮 on 𝑌
by 𝑓∗ 𝒮 (𝑈 ) = 𝒮 (𝑓 −1 (𝑈 )); and define restriction maps 𝑟∗ 𝑈
𝑉 ∶ 𝑓∗ 𝒮 (𝑈 ) →
𝑓∗ 𝒮 (𝑉 ) by 𝑟∗ 𝑈
𝑉 = 𝑟𝑓 −1 (𝑉 ) .
𝑓 −1 (𝑈 )

(a) Show that 𝑓∗ 𝒮 is a sheaf of abelian groups on 𝑌 , called a direct image

(b) Show that for each sheaf morphism 𝐴 ∶ 𝒮 → 𝒮 ′ between sheaves on

sheaf .

𝑋, there is a morphism 𝐴∗ ∶ 𝑓∗ 𝒮 → 𝑓∗ 𝒮 ′ , so that the assignment

𝒮 ↦ 𝑓∗ 𝒮 , 𝐴 ↦ 𝐴∗ is a covariant functor from the category of
sheaves of abelian groups on 𝑋 to the analogous category on 𝑌 .
(c) Show that if 𝑓 is a constant map, then 𝑓∗ 𝒮 is a skyscraper sheaf
whose nontrivial stalk is the space 𝒮 (𝑋) of global sections.
5-13. Let 𝑋, 𝑌 , and 𝑓 be as in Problem 5-12. In addition to the operation taking
sheaves on 𝑋 to direct image sheaves on 𝑌 , there is a reverse operation
taking sheaves on 𝑌 to sheaves on 𝑋, but it is a little less straightforward
to define. Given a sheaf 𝒯 of abelian groups on 𝑌 , let 𝜋 ∶ Et(𝒯 ) → 𝑌
be its étalé space, let 𝒯 # ⊆ 𝑋 × Et(𝒯 ) be the following fiber product:
𝒯 # = {(𝑥, 𝜏) ∈ 𝑋 × Et(𝒯 ) ∶ 𝑓 (𝑥) = 𝜋(𝜏)}.
(a) Show that with the subspace topology and the projection 𝜋1 ∶ 𝒯 # →
𝑋 given by 𝜋1 (𝑥, 𝜏) = 𝑥, 𝒯 # is an étalé space of abelian groups on
𝑋. Its sheaf of sections, denoted by 𝑓 −1 𝒯 , is called an inverse image

(b) Show that for each sheaf morphism 𝐴 ∶ 𝒯 → 𝒯 ′ between sheaves

sheaf .

on 𝑌 , it is possible to assign a morphism 𝑓 −1 𝐴 ∶ 𝑓 −1 𝒯 → 𝑓 −1 𝒯 ′

in such a way as to define a covariant functor from sheaves on 𝑌 to
sheaves on 𝑋.
(c) Show that if 𝑓 ∶ 𝑈 → 𝑌 is inclusion of an open subset, then 𝑓 −1 𝒯
is isomorphic to the restriction 𝒯 |𝑈 defined by (5.3).
(d) Show that if 𝑓 ∶ 𝑋 → 𝑌 is a constant map with value 𝑐 ∈ 𝑌 , then
𝑓 −1 𝒯 is the constant sheaf 𝒯𝑐 , where 𝒯𝑐 is the stalk of 𝒯 at 𝑐.
 10.1090/gsm/244/06

Chapter 6

Sheaf Cohomology

As the discussion in the previous chapter suggests, an important question in the the-
ory of sheaves is ascertaining when the global section sequence associated with an
exact sheaf sequence is exact, and if not, how to characterize the failure of exactness.
In this chapter, we introduce a powerful machine called sheaf cohomology that can
answer questions like these and many more. There is a considerable amount of tech-
nical work that has to be done to establish the necessary results; but the work will
pay off richly when we start seeing applications of the theory to complex manifolds.

Definitions
As with many aspects of this subject, there are various definitions of sheaf coho-
mology available. The construction we will give is called Čech cohomology after
the early twentieth century Czech mathematician Eduard Čech, who introduced the
main ideas behind it in 1932 [Čec32]. It is not the most general construction, be-
cause it only behaves well on paracompact Hausdorff spaces; but that topological
restriction is not an issue for us because we only need to consider sheaves on man-
ifolds, and the Čech construction is supremely well suited to our purposes. At the
end of the chapter, we will explain its relationship with other sheaf cohomology

Suppose 𝑀 is a topological space and 𝒮 is a sheaf of abelian groups on 𝑀.

theories.

Given an indexed open cover 𝒰 = {𝑈𝛼 }𝛼∈𝐴 of 𝑀 and a nonnegative integer 𝑝,

a 𝒑-cochain on 𝓤 with coefficients in 𝓢 (sometimes called a Čech cochain to

𝑐 that assigns to every multi-index (𝛼0 , … , 𝛼𝑝 ) of length (𝑝 + 1) a section 𝑐𝛼0 …𝛼𝑝 ∈

distinguish it from the singular cochains used in algebraic topology) is an operator

𝒮 (𝑈𝛼0 ∩⋯∩𝑈𝛼𝑝 ). (Note that this means 𝑐𝛼0 …𝛼𝑝 = 0 whenever 𝑈𝛼0 ∩⋯∩𝑈𝛼𝑝 = ∅.)
The 𝒑th cochain group on 𝓤 with coefficients in 𝓢 is the set 𝐶 𝑝 (𝒰; 𝒮 ) of all such

147
148 6. Sheaf Cohomology

cochains, which is an abelian group under the addition operation defined by

(𝑐 + 𝑐 ′ )𝛼0 …𝛼𝑝 = 𝑐𝛼0 …𝛼𝑝 + 𝑐𝛼′ 0 …𝛼𝑝 .

If 𝒮 is a sheaf of real or complex vector spaces, then 𝐶 𝑝 (𝒰; 𝒮 ) is also a vector space
with the obvious scalar multiplication. (This notation 𝐶 𝑝 (𝒰; 𝒮 ), with arguments

𝐶 𝑝 (𝑀, 𝑁) for spaces of maps of class 𝐶 𝑝 between manifolds.)

consisting of an open cover and a sheaf, should not be confused with notations like

The coboundary operator is the homomorphism 𝛿 ∶ 𝐶 𝑝 (𝒰; 𝒮 ) → 𝐶 𝑝+1 (𝒰; 𝒮 )

defined by
𝑝+1
(𝛿𝑐)𝛼0 …𝛼𝑝+1 = (−1)𝑗 𝑐𝛼0 …𝛼̂𝑗 …𝛼𝑝+1 | .
∑ 𝑈0 ∩⋯∩𝑈𝑝+1
𝑗=0

Lemma 6.1. With 𝛿 defined as above, 𝛿 ∘ 𝛿 = 0.

Proof. We compute

(𝛿𝛿𝑐)𝛼0 …𝛼𝑝+2 = (−1)𝑗 (𝛿𝑐)𝛼0 …𝛼̂𝑗 …𝛼𝑝+2 |

∑ 𝑈0 ∩⋯∩𝑈𝑝+2
𝑗
0≤𝑗≤𝑝+1

= (−1)𝑗+𝑘 𝑐𝛼0 …̂
𝛼𝑘 …𝛼̂𝑗 …𝛼𝑝+1 |
∑ 𝑈0 ∩⋯∩𝑈𝑝+2
𝑗,𝑘
0≤𝑘<𝑗≤𝑝+1

+ (−1)𝑗+𝑘−1 𝑐𝛼0 …𝛼̂𝑗 …̂

𝛼𝑘 …𝛼𝑝+1 | ,
∑ 𝑈0 ∩⋯∩𝑈𝑝+2
𝑗,𝑘
0≤𝑗<𝑘≤𝑝+1

where the sign in the last sum reflects the fact that when 𝑘 > 𝑗, the index 𝛼𝑘 is in
position 𝑘 − 1 in (𝛿𝑐)𝛼0 …𝛼̂𝑗 …𝛼𝑝+2 . After interchanging the dummy indices 𝑗 and 𝑘
in the last sum, we see that these two sums exactly cancel each other. □

(Those who have studied simplicial cohomology might recognize these for-
mulas as being close kin to the formula for the coboundary operator in simplicial
cohomology. This connection is explained in Problem 6-10.)
Lemma 6.1 shows that the cochain groups fit together in a cochain complex:
𝛿 𝛿 𝛿 𝛿
0 → 𝐶 0 (𝒰; 𝒮 ) ⟶ 𝐶 1 (𝒰; 𝒮 ) ⟶ ⋯ ⟶ 𝐶 𝑝 (𝒰; 𝒮 ) ⟶ 𝐶 𝑝+1 (𝒰; 𝒮 ) → ⋯ .

on 𝓤 with coefficients in 𝓢:
The cohomology groups of this complex are called the Čech cohomology groups

Ker (𝛿 ∶ 𝐶 𝑝 (𝒰; 𝒮 ) → 𝐶 𝑝+1 (𝒰; 𝒮 ))

𝐻 𝑝 (𝒰; 𝒮 ) =
Im (𝛿 ∶ 𝐶 𝑝−1 (𝒰; 𝒮 ) → 𝐶 𝑝 (𝒰; 𝒮 ))
Definitions 149

(where we interpret 𝐶 𝑝 (𝒰; 𝒮 ) to be zero when 𝑝 < 0). In general these are abelian
groups; if 𝒮 is a sheaf of (real or complex) vector spaces, then each 𝐻 𝑝 (𝒰; 𝒮 ) is a

A cochain 𝑐 ∈ 𝐶 𝑝 (𝒰; 𝒮 ) satisfying 𝛿𝑐 = 0 is called a (Čech) cocycle, and

vector space.

one satisfying 𝑐 = 𝛿𝑏 for some 𝑏 ∈ 𝐶 𝑝−1 (𝒰; 𝒮 ) is called a (Čech) coboundary.

We denote the group of cocycles on 𝒰 by 𝑍 𝑝 (𝒰; 𝒮 ), and the group of cobound-
aries by 𝐵 𝑝 (𝒰; 𝒮 ), so we can also write 𝐻 𝑝 (𝒰; 𝒮 ) = 𝑍 𝑝 (𝒰; 𝒮 )/𝐵 𝑝 (𝒰; 𝒮 ). The
equivalence class of a cocycle 𝑐 in this quotient is denoted by [𝑐] and is called a
cohomology class; two cocycles that determine the same cohomology class are said
to be cohomologous.
This is already rather abstract, so let us look at a few examples before proceed-
ing.
Example 6.2 (Čech Cocycles in Degree 0). Suppose 𝒮 is a sheaf of abelian groups
on a topological space 𝑀 and 𝒰 = {𝑈𝛼 }𝛼∈𝐴 is an indexed open cover of 𝑀. A
0-cochain 𝑐 on 𝒰 just assigns a section 𝑐𝛼 ∈ 𝒮 (𝑈𝛼 ) for each 𝛼. The coboundary of

(𝛿𝑐)𝛼𝛽 = 𝑐𝛽 |𝑈 ∩𝑈 − 𝑐𝛼 |𝑈 ∩𝑈 .
such a cochain is
𝛼 𝛽 𝛼 𝛽
Since there are no (−1)-cochains, there are no nontrivial 0-coboundaries, and thus
𝐻 0 (𝒰; 𝒮 ) is equal to the space 𝑍 0 (𝒰; 𝒮 ) of cocycles.
One way to obtain such a cocycle is to start with a global section 𝜎 ∈ 𝒮 (𝑀),
and define a 0-cochain 𝑐 by 𝑐𝛼 = 𝜎|𝑈𝛼 for each 𝛼. It follows automatically from
the properties of the sheaf restriction maps that 𝑐 is a cocycle, so this defines a
map 𝐼𝒰,𝒮 ∶ 𝒮 (𝑀) → 𝐻 0 (𝒰; 𝒮 ), which is a homomorphism of whatever algebraic
structures 𝒮 (𝑀) and 𝐻 0 (𝒰; 𝒮 ) are endowed with. Injectivity of 𝐼𝒰,𝒮 follows from
the locality property of sheaves, and surjectivity from the gluing property, so 𝐼𝒰,𝒮
is an isomorphism. Thus 𝐻 0 (𝒰; 𝒮 ) is always isomorphic to the space 𝒮 (𝑀) of
global sections of 𝒮 . //
Example 6.3 (Čech Cocycles in Degree 1). Let 𝑀, 𝒮 , and 𝒰 be as in the preceding
example. A 1-cochain on 𝒰 is a choice of a section 𝑐𝛼𝛽 ∈ 𝒮 (𝑈𝛼 ∩ 𝑈𝛽 ) for each pair
of indices 𝛼, 𝛽. The coboundary operator is
(𝛿𝑐)𝛼𝛽𝛾 = 𝑐𝛽𝛾 − 𝑐𝛼𝛾 + 𝑐𝛼𝛽 on 𝑈𝛼 ∩ 𝑈𝛽 ∩ 𝑈𝛾
(where the phrase “on 𝑈𝛼 ∩𝑈𝛽 ∩𝑈𝛾 ” should be interpreted to mean that each section
is restricted to that set before performing the addition and subtraction). Thus 𝑐 is
a cocycle if and only if 𝑐𝛼𝛾 = 𝑐𝛼𝛽 + 𝑐𝛽𝛾 on 𝑈𝛼 ∩ 𝑈𝛽 ∩ 𝑈𝛾 for all 𝛼, 𝛽, 𝛾. Two such
cocycles 𝑐 and 𝑐 ′ are cohomologous if and only if there is a 0-cochain 𝑏 such that
𝑐𝛼𝛽 − 𝑐𝛼𝛽
′
= 𝑏 𝛽 − 𝑏𝛼 on 𝑈𝛼 ∩ 𝑈𝛽 .

If 𝒮 is a sheaf of abelian groups written multiplicatively, we would write

(𝛿𝑐)𝛼𝛽𝛾 = 𝑐𝛽𝛾 𝑐𝛼𝛾
−1
𝑐𝛼𝛽 on 𝑈𝛼 ∩ 𝑈𝛽 ∩ 𝑈𝛾 ,
150 6. Sheaf Cohomology

so the formula for 𝑐 to be a cocycle would be

𝑐𝛼𝛽 𝑐𝛽𝛾 = 𝑐𝛼𝛾 on 𝑈𝛼 ∩ 𝑈𝛽 ∩ 𝑈𝛾 .
Two 1-cocycles 𝑐 and 𝑐 are cohomologous if and only if there is a 0-cochain 𝑏 such
′

𝑐𝛼𝛽 = 𝑐𝛼𝛽 𝑏𝛽 𝑏−1 on 𝑈𝛼 ∩ 𝑈𝛽 .

that
′
𝛼
This might look remarkably similar to the isomorphism criterion for line bundles
in terms of their transition functions; that is not an accident, as we will see later in
this chapter. //
Example 6.4 (1-Cocycles of Closed Forms). Suppose 𝑀 is a smooth manifold
and 𝜔 is a smooth closed 𝑘-form on 𝑀. By the Poincaré lemma, we can find an
open cover 𝒰 = {𝑈𝛼 }𝛼∈𝐴 such that on each 𝑈𝛼 there is a smooth (𝑘 − 1)-form 𝜂𝛼
satisfying 𝜔|𝑈𝛼 = 𝑑𝜂𝛼 . These (𝑘 − 1)-forms might not agree where they overlap,
so for each 𝛼 and 𝛽 we can define a (𝑘 − 1)-form 𝛾𝛼𝛽 on 𝑈𝛼 ∩ 𝑈𝛽 by
𝛾𝛼𝛽 = 𝜂𝛽 |𝑈 − 𝜂 𝛼 |𝑈 .
𝛼 ∩𝑈𝛽 𝛼 ∩𝑈𝛽

Since 𝑑𝜂𝛽 = 𝜔 = 𝑑𝜂𝛼 on the intersection of their domains, 𝛾𝛼𝛽 is a closed form,
and thus this assignment defines a 1-cochain 𝛾 ∈ 𝐶 1 (𝒰; 𝒵 𝑘−1 ), where as before
𝒵 𝑘−1 is the sheaf of closed (𝑘 − 1)-forms on 𝑀. It follows immediately from
the definition of 𝛾 that 𝛿𝛾 = 0, so 𝛾 is actually a 1-cocycle with coefficients in
𝒵 𝑘−1 . It is a coboundary if and only if there is a collection of closed (𝑘 − 1)-forms
𝜎𝛼 ∈ 𝒵 𝑘−1 (𝑈𝛼 ) such that
𝛾𝛼𝛽 = 𝜎𝛽 − 𝜎𝛼 on 𝑈𝛼 ∩ 𝑈𝛽 ,

𝜎𝛽 − 𝜎𝛼 = 𝜂𝛽 − 𝜂𝛼 on 𝑈𝛼 ∩ 𝑈𝛽 .
which is to say

If this is the case, then we see that 𝜂𝛼 − 𝜎𝛼 and 𝜂𝛽 − 𝜎𝛽 restrict to the same form on
the intersection of their domains, so they piece together to produce a global (𝑘 − 1)-
form 𝜃 such that 𝑑𝜃 = 𝜔 (because 𝑑𝜃 = 𝑑𝜂𝛼 − 𝑑𝜎𝛼 = 𝜔 − 0 on 𝑈𝛼 ). Conversely,
if 𝜔 is exact, we can choose the 𝜂𝛼 ’s all to be restrictions of a global form 𝜂, so
the cocycle 𝛾 is zero. Thus starting with a closed 𝑘-form 𝜔, we have produced a
1-cocycle on 𝒰 with coefficients in 𝒵 𝑘−1 , which is a coboundary if and only if 𝜔
is exact. This is a special case of a deep connection between de Rham cohomology
and sheaf cohomology, which we will explore later in the chapter. //

Dependence on the Open Cover

choice of open cover. Given an indexed open cover 𝒰 = {𝑈𝛼 }𝛼∈𝐴 , recall that
Next we have to examine how the Čech cohomology groups depend on the

another such open cover 𝒱 = {𝑉𝛽 }𝛽∈𝐵 is a refinement of 𝓤 if for each 𝛽 ∈ 𝐵, there
is some 𝛼 ∈ 𝐴 such that 𝑉𝛽 ⊆ 𝑈𝛼 . If this is the case, then we can choose a refining
map 𝜌 ∶ 𝒱 → 𝒰, by which we mean a map 𝜌 ∶ 𝐵 → 𝐴 such that 𝑉𝛽 ⊆ 𝑈𝜌(𝛽) for
Definitions 151

every 𝛽. Using this, we define a homomorphism 𝜌# ∶ 𝐶 𝑝 (𝒰; 𝒮 ) → 𝐶 𝑝 (𝒱 ; 𝒮 ) for

each 𝑝 by
(𝜌# 𝑐)𝛽0 …𝛽𝑝 = 𝑐𝜌(𝛽0 )…𝜌(𝛽𝑝 ) |𝑉 .
𝛽0 ∩⋯∩𝑉𝛽𝑝

Then a straightforward computation shows that 𝛿(𝜌# 𝑐) = 𝜌# (𝛿𝑐), so 𝜌# maps cocy-

morphism 𝜌∗ ∶ 𝐻 𝑝 (𝒰; 𝒮 ) → 𝐻 𝑝 (𝒱 ; 𝒮 ). If 𝒮 is a sheaf of real or complex vector

cles to cocycles and coboundaries to coboundaries, and thus descends to a homo-

spaces, then 𝜌∗ is linear.

Lemma 6.5. The homomorphism 𝜌∗ is independent of the choice of refining map.

Proof. Given 𝒰 and 𝒱 as above, suppose 𝜌, 𝜌 ̃ ∶ 𝒱 → 𝒰 are two different refining

maps. For each 𝑝, define a homomorphism 𝜃 ∶ 𝐶 𝑝 (𝒰; 𝒮 ) → 𝐶 𝑝−1 (𝒱 ; 𝒮 ) by
𝑝−1
(𝜃𝑐)𝛽0 …𝛽𝑝−1 = (−1)𝑗 𝑐𝜌(𝛽0 )…𝜌(𝛽𝑗 )𝜌(𝛽 ̃ 𝑝−1 ) |𝑉 .
∑ ̃ 𝑗 )…𝜌(𝛽 𝛽0 ∩⋯∩𝑉𝛽𝑝−1
𝑗=0

We will show that 𝜃 satisfies the following cochain homotopy formula:

(6.1) 𝜃𝛿 + 𝛿𝜃 = 𝜌#̃ − 𝜌# .

(The terminology comes from algebraic topology, where such formulas are com-
monly used for showing that two maps between chain complexes or cochain com-
plexes induce the same map on homology or cohomology; a classic example is to
show that homotopic maps between topological spaces induce the same homology

Granting (6.1) for now, given a cohomology class [𝑐] ∈ 𝐻 𝑝 (𝒰; 𝒮 ) represented
and cohomology homomorphisms.)

by a cocycle 𝑐 ∈ 𝑍 𝑝 (𝒰; 𝒮 ), we see that 𝜌#̃ 𝑐 − 𝜌# 𝑐 = 0 + 𝛿𝜃𝑐, so 𝜌∗̃ [𝑐] = [𝜌#̃ 𝑐] =

[𝜌# 𝑐] = 𝜌∗ [𝑐], which proves the lemma.
To complete the proof, we need to verify the cochain homotopy formula (6.1).
This is just a messy computation. Here it is. (We suppress the restrictions from the

restricted to 𝑉𝛽0 ∩ ⋯ ∩ 𝑉𝛽𝑝 .)

notation for brevity; all of the sections on the right-hand side are understood to be

𝑝
(𝜃𝛿𝑐)𝛽0 …𝛽𝑝 = (−1)𝑗 (𝛿𝑐)𝜌(𝛽0 )…𝜌(𝛽𝑗 )𝜌(𝛽
∑ ̃ 𝑗 )…𝜌(𝛽
̃ 𝑝)
𝑗=0

= (−1)𝑗+𝑘 𝑐𝜌(𝛽
∑ ̂
0 )…𝜌(𝛽𝑘 )…𝜌(𝛽𝑗 )𝜌(𝛽
̃ 𝑗 )…𝜌(𝛽
̃ 𝑝)
𝑗,𝑘
0≤𝑘≤𝑗≤𝑝

+ (−1)𝑗+𝑘+1 𝑐𝜌(𝛽 ;
∑ 0 )…𝜌(𝛽𝑗 )𝜌(𝛽
̃ 𝑗 )…𝜌(𝛽̂
̃ 𝑘 )…𝜌(𝛽
̃ 𝑝)
𝑗,𝑘
0≤𝑗≤𝑘≤𝑝
152 6. Sheaf Cohomology

𝑝
(𝛿𝜃𝑐)𝛽0 …𝛽𝑝 = (−1)𝑘 (𝜃𝑐)𝛽
∑ ̂
0 …𝛽𝑘 …𝛽𝑝
𝑘=0

= (−1)𝑘+𝑗 𝑐𝜌(𝛽
∑ 0 )…𝜌(𝛽𝑗 )𝜌(𝛽
̂
̃ 𝑗 )…𝜌(𝛽
̃ 𝑘 )…𝜌(𝛽
̃ 𝑝)
𝑗,𝑘
0≤𝑗<𝑘≤𝑝

+ (−1)𝑘+𝑗−1 𝑐𝜌(𝛽 .
∑ ̂
0 )…𝜌(𝛽𝑘 )…𝜌(𝛽𝑗 )𝜌(𝛽
̃ 𝑗 )…𝜌(𝛽
̃ 𝑝)
𝑗,𝑘
0≤𝑘<𝑗≤𝑝

When these expressions are added together, each term in the expression for 𝜃𝛿𝑐
is canceled by one term in the expression for 𝛿𝜃𝑐, except the terms in 𝜃𝛿𝑐 where
𝑗 = 𝑘. Those terms give
𝑝
(𝜃𝛿𝑐 + 𝛿𝜃𝑐)𝛽0 …𝛽𝑝 = 𝑐𝜌(𝛽0 )…𝜌(𝛽𝑗−1 )𝜌(𝛽
∑ ̃ 𝑗 )…𝜌(𝛽
̃ 𝑝)
𝑗=0
𝑝
+ (−1)𝑐𝜌(𝛽0 )…𝜌(𝛽𝑗 )𝜌(𝛽 ̃ 𝑝).
∑ ̃ 𝑗+1 )…𝜌(𝛽
𝑗=0

This is a telescoping sum in which all the terms cancel except the 𝑗 = 0 term in the
first sum and the 𝑗 = 𝑝 term in the second sum; what is left is exactly (𝜌#̃ 𝑐)𝛽0 …𝛽𝑝 −
(𝜌# 𝑐)𝛽0 …𝛽𝑝 . □

The result of this lemma is that whenever 𝒱 is a refinement of 𝒰, there is a

canonical homomorphism 𝜌∗𝒰𝒱 ∶ 𝐻 𝑝 (𝒰; 𝒮 ) → 𝐻 𝑝 (𝒱 ; 𝒮 ). Moreover, if in addi-
tion 𝒲 is a refinement of 𝒱 , we can choose our refining maps so that the refining
map from 𝒲 to 𝒰 is the composition of the maps from 𝒲 to 𝒱 and 𝒱 to 𝒰, so
that 𝜌∗𝒱 𝒲 ∘ 𝜌∗𝒰𝒱 = 𝜌∗𝒰𝒲 . It follows that the collection of all open covers of 𝑀
with the homomorphisms 𝜌∗𝒰𝒱 is a direct system—given any two open covers, the

ers with the relation 𝒰 ≤ 𝒱 if 𝒱 is a refinement of 𝒰 is an example of a directed

collection of all of their intersections is a refinement of both. (The set of open cov-

set that is not partially ordered: two open covers can be refinements of each other

We define the 𝒑th sheaf cohomology group of 𝑴 with coefficients in 𝓢 as the

without being equal.)

direct limit of this system:

𝐻 𝑝 (𝑀; 𝒮 ) = lim 𝐻 𝑝 (𝒰; 𝒮 ).

⟶

If 𝒮 is a sheaf of real or complex vector spaces, then 𝐻 𝑝 (𝑀; 𝒮 ) is a vector space.

in 𝓢, to distinguish them from other constructions of sheaf cohomology groups; see

These groups are sometimes called the Čech cohomology groups with coefficients

the last section of this chapter for the relationships among the various constructions.
Definitions 153

For each open cover 𝒰, there is a canonical homomorphism from 𝐻 𝑝 (𝒰; 𝒮 )

to 𝐻 𝑝 (𝑀; 𝒮 ), which sends a cohomology class [𝑐] to its equivalence class in the
direct limit. Let us denote this equivalence class by [[𝑐]] ∈ 𝐻 𝑝 (𝑀; 𝒮 ). Unwinding
the definitions, we see that for a given cover 𝒰 = {𝑈𝛼 }𝛼∈𝐴 , two cocycles 𝑐, 𝑐 ′ ∈
𝐶 𝑝 (𝒰; 𝒮 ) satisfy [[𝑐]] = [[𝑐 ′ ]] if and only if there is a cover 𝒱 = {𝑉𝛽 }𝛽∈𝐵 refining
𝒰 and a refining map 𝜌 ∶ 𝒱 → 𝒰 such that 𝜌# (𝑐 −𝑐 ′ ) is a coboundary in 𝐶 𝑝 (𝒱 ; 𝒮 ).
Theorem 6.6 (Functoriality of Sheaf Cohomology). Suppose 𝒮 , 𝒯 are sheaves
of abelian groups on a topological space 𝑀 and 𝐹 ∶ 𝒮 → 𝒯 is a sheaf morphism.
For each 𝑝, there is a homomorphism 𝐹∗ ∶ 𝐻 𝑝 (𝑀; 𝒮 ) → 𝐻 𝑝 (𝑀; 𝒯 ), and these
homomorphisms satisfy
(Id𝒮 )∗ = Id𝐻 𝑝 (𝑀;𝒮 ) and (𝐹 ∘ 𝐺)∗ = 𝐹∗ ∘ 𝐺∗ .
If 𝐹 is a morphism between sheaves of real or complex vector spaces, then 𝐹∗ is
linear.

Proof. For a specific open cover 𝒰 = {𝑈𝛼 }, define 𝐹# ∶ 𝐶 𝑝 (𝒰; 𝒮 ) → 𝐶 𝑝 (𝒰; 𝒯 )

by
(6.2) (𝐹# 𝑐)𝛼0 …𝛼𝑝 = 𝐹 (𝑐𝛼0 …𝛼𝑝 ).
It clearly satisfies Id# = Id and (𝐹 ∘ 𝐺)# = 𝐹# ∘ 𝐺# . Moreover, a straightforward
computation shows that
(6.3) 𝐹# ∘ 𝛿 = 𝛿 ∘ 𝐹 # ,
which implies that 𝐹# descends to a homomorphism (still denoted by 𝐹# ) from
𝐻 𝑝 (𝒰; 𝒮 ) to 𝐻 𝑝 (𝒰; 𝒯 ). Another straightforward computation shows that for any
pair of covers 𝒰 and 𝒱 equipped with a refining map 𝜌 ∶ 𝒱 → 𝒰, we have 𝐹# ∘
𝜌# = 𝜌# ∘ 𝐹# , so 𝐹# descends to a homomorphism 𝐹∗ ∶ 𝐻 𝑝 (𝑀; 𝒮 ) → 𝐻 𝑝 (𝑀; 𝒯 )
satisfying the conclusions of the proposition. For sheaves of vector spaces, one
need only check that all of the homomorphisms above are linear maps. □
Theorem 6.7 (Degree 0 Sheaf Cohomology). For every sheaf 𝒮 of abelian groups
on a topological space 𝑀, there is an isomorphism 𝐼𝒮 ∶ 𝒮 (𝑀) ≅ 𝐻 0 (𝑀; 𝒮 ) such
that for every sheaf morphism 𝐹 ∶ 𝒮 → 𝒯 , the following diagram commutes:
𝐼𝒮
𝒮 (𝑀) / 𝐻 0 (𝑀; 𝒮 )

𝐹𝑀 𝐹∗
𝐼𝒯
(6.4)

𝒯 (𝑀) / 𝐻 0 (𝑀; 𝒯 ).
 

If 𝒮 is a sheaf of real or complex vector spaces, then 𝐼𝒮 is linear.

Proof. Suppose 𝒮 is a sheaf of abelian groups on 𝑀. Example 6.2 showed that

for each open cover 𝒰 of 𝑀, there is an isomorphism 𝐼𝒰,𝒮 ∶ 𝒮 (𝑀) → 𝐻 0 (𝒰; 𝒮 ).
154 6. Sheaf Cohomology

For any open cover 𝒱 refining 𝒰, it follows from the definitions that these isomor-
phisms commute with refining maps: 𝜌∗𝒰𝒱 ∘ 𝐼𝒰,𝒮 = 𝐼𝒱 ,𝒮 . Thus they pass to the
direct limit to yield a canonical isomorphism 𝐼𝒮 ∶ 𝒮 (𝑀) → 𝐻 0 (𝑀; 𝒮 ), which is
linear if 𝒮 is a sheaf of vector spaces.
If 𝐹 ∶ 𝒮 → 𝒯 is a sheaf morphism, recall that for each open cover 𝒰, the map
𝐹# ∶ 𝐻 0 (𝒰; 𝒮 ) → 𝐻 0 (𝒰; 𝒯 ) is defined by (𝐹# 𝑐)𝛼 = 𝐹 (𝑐𝛼 ). It is immediate that
𝐹# ∘ 𝐼𝒰,𝒮 = 𝐼𝒰,𝒯 ∘ 𝐹𝑀 , and then commutativity of (6.4) follows by passing to the
direct limit. □

The Long Exact Cohomology Sequence

By far the most important property of the sheaf cohomology groups is expressed in

If 𝐴∗ , 𝐵 ∗ , and 𝐶 ∗ are cochain complexes, a sequence of cochain maps

the next theorem. It relies on the following standard result in homological algebra.

𝛼 𝛽
(6.5) 0 → 𝐴∗ → 𝐵 ∗ → 𝐶 ∗ → 0

phisms 0 → 𝐴𝑝 → 𝐵 𝑝 → 𝐶 𝑝 → 0 is exact.
is called an exact sequence of cochain complexes if each sequence of homomor-

cochain complexes of the form (6.5). Then for each 𝑝 there is a connecting homo-
Lemma 6.8 (The Zigzag Lemma). Suppose we are given an exact sequence of

morphism 𝛿∗ ∶ 𝐻 𝑝 (𝐶 ∗ ) → 𝐻 𝑝+1 (𝐴∗ ) such that the following sequence is exact:

𝛿∗ 𝛼∗ 𝛽∗ 𝛿∗ 𝛼∗
⋯ ⟶ 𝐻 𝑝 (𝐴∗ ) ⟶ 𝐻 𝑝 (𝐵 ∗ ) ⟶ 𝐻 𝑝 (𝐶 ∗ ) ⟶ 𝐻 𝑝+1 (𝐴∗ ) ⟶ ⋯ .
For each 𝑝, the homomorphism 𝛿∗ ∶ 𝐻 𝑝 (𝐶 ∗ ) → 𝐻 𝑝+1 (𝐴∗ ) is characterized as fol-
lows:
𝛿∗ ([𝑐]) = [𝑎] if and only if there exists some 𝑏 ∈ 𝐵 𝑝 such that
𝛽(𝑏) = 𝑐 and 𝛼(𝑎) = 𝛿𝑏.
(6.6)

If all of the groups and homomorphisms are real or complex vector spaces, then the
connecting homomorphisms are linear maps.
See [LeeTM, Lemma 13.17] or [Hat02, Thm. 2.16] for a proof. (The zigzag
lemma is stated and proved there for chain complexes, in which the arrows go in the
direction of decreasing indices, but the proof for cochain complexes works exactly
the same way.)
Theorem 6.9 (The Long Exact Sequence in Sheaf Cohomology). Suppose 𝒜 ,
ℬ, and 𝒞 are sheaves of abelian groups on a paracompact Hausdorff space 𝑀,
and the following sequence of sheaf morphisms is exact:
𝛼 𝛽
(6.7) 0 → 𝒜 ⟶ ℬ ⟶ 𝒞 → 0.
The Long Exact Cohomology Sequence 155

Then for each 𝑝 ≥ 0, there exists a connecting homomorphism

𝛿∗ ∶ 𝐻 𝑝 (𝑀; 𝒞 ) → 𝐻 𝑝+1 (𝑀; 𝒜 )

such that the following sequence is exact:

𝛼∗ 𝛽∗ 𝛿∗ 𝛼∗
(6.8) 0 → 𝐻 0 (𝑀; 𝒜 ) ⟶ 𝐻 0 (𝑀; ℬ) ⟶ 𝐻 0 (𝑀; 𝒞 ) ⟶ 𝐻 1 (𝑀; 𝒜 ) ⟶ ⋯
𝛼∗ 𝛽∗ 𝛿∗ 𝛼∗
⋯ ⟶ 𝐻 𝑝 (𝑀; ℬ) ⟶ 𝐻 𝑝 (𝑀; 𝒞 ) ⟶ 𝐻 𝑝+1 (𝑀; 𝒜 ) ⟶ ⋯ .

If 𝒜 , ℬ, and 𝒞 are all sheaves of real or complex vector spaces, then 𝛿∗ is linear. It
satisfies the following naturality property: given a commutative diagram of sheaves
and sheaf morphisms

𝛼 𝛽
0 /𝒜 /ℬ /𝒞 /0

(6.9) 𝐴 𝐵 𝐶
0 / 𝒜′ / ℬ′ / 𝒞′ /0
𝛼′ 𝛽′
  

in which the horizontal rows are exact, the two connecting homomorphisms 𝛿∗ and
𝛿∗′ satisfy 𝐴∗ ∘ 𝛿∗ = 𝛿∗′ ∘ 𝐶∗ for each 𝑝:

𝛿∗
𝐻 𝑝 (𝑀; 𝒞 ) / 𝐻 𝑝+1 (𝑀; 𝒜 )

(6.10) 𝐶∗ 𝐴∗

𝐻 𝑝 (𝑀; 𝒞 ′ ) / 𝐻 𝑝+1 (𝑀; 𝒜 ′ ).

𝛿∗′
 

Proof. Begin with an arbitrary open cover 𝒰 for 𝑀. For each 𝑝 ≥ 0, consider the
following sequence of cochain groups:

𝛼# 𝛽#
0 → 𝐶 𝑝 (𝒰; 𝒜 ) ⟶ 𝐶 𝑝 (𝒰; ℬ) ⟶ 𝐶 𝑝 (𝒰; 𝒞 ).

Proposition 5.24 applied on each intersection 𝑈𝛼0 ∩⋯∩𝑈𝛼𝑝 shows that this sequence
is exact. However, 𝛽# might not be surjective. So we cheat: define the subgroup
𝐶𝛽 (𝒰; 𝒞 ) ⊆ 𝐶 𝑝 (𝒰; 𝒞 ) to be the image of 𝛽# ∶ 𝐶 𝑝 (𝒰; ℬ) → 𝐶 𝑝 (𝒰; 𝒞 ), so we have
𝑝

a short exact sequence

𝛼# 𝛽#
0 → 𝐶 𝑝 (𝒰; 𝒜 ) ⟶ 𝐶 𝑝 (𝒰; ℬ) ⟶ 𝐶𝛽 (𝒰; 𝒞 ) → 0.
𝑝
156 6. Sheaf Cohomology

sider cochain groups to be zero in degrees less than 0):

Now consider the following diagram of group homomorphisms (where we con-

𝛼# 𝛽#
0 / 𝐶 𝑝−1 (𝒰; 𝒜 ) / 𝐶 𝑝−1 (𝒰; ℬ) / 𝐶 𝑝−1 (𝒰; 𝒞 )
𝛽
/0

𝛿 𝛿 𝛿
𝛼# 𝛽#
0 / 𝐶 𝑝 (𝒰; 𝒜 ) / 𝐶 𝑝 (𝒰; ℬ) / 𝐶 𝑝 (𝒰; 𝒞 )
𝛽
/0
  

𝛿 𝛿 𝛿
𝛼# 𝛽#
0 / 𝐶 𝑝+1 (𝒰; 𝒜 ) / 𝐶 𝑝+1 (𝒰; ℬ) / 𝐶 𝑝+1 (𝒰; 𝒞 ) / 0.

𝛽
 

Since 𝛿 commutes with 𝛽# by (6.3), it follows that 𝛿 takes 𝐶𝛽 (𝒰; 𝒞 ) to 𝐶𝛽 (𝒰; 𝒞 )

𝑝−1 𝑝

and 𝐶𝛽 (𝒰; 𝒞 ) to 𝐶𝛽 (𝒰; 𝒞 ). The horizontal rows of this diagram are exact, and
𝑝 𝑝+1

the columns are cochain complexes; and it commutes by (6.3). Let 𝐻𝛽∗ (𝒰; 𝒞 ) de-
note the cohomology of the cochain complex 𝐶𝛽∗ (𝒰; 𝒞 ).
The zigzag lemma shows that for each 𝑝 there is a connecting homomorphism
𝛿∗ ∶ 𝐻𝛽 (𝒰; 𝒞 ) → 𝐻 𝑝+1 (𝒰; 𝒜 ) such that the following sequence is exact:
𝑝

𝛼∗ 𝛽∗ 𝛿∗
⋯ → 𝐻 𝑝 (𝒰; 𝒜 ) ⟶ 𝐻 𝑝 (𝒰; ℬ) ⟶ 𝐻𝛽 (𝒰; 𝒞 ) ⟶ 𝐻 𝑝+1 (𝒰; 𝒜 ) → ⋯ .
𝑝

a cover 𝒱 refining 𝒰 and a refining map 𝜌 ∶ 𝒱 → 𝒰, we get a homomorphism

Next we need to examine what happens when we pass to the direct limit. Given

𝜌# ∶ 𝐻 𝑝 (𝒰; 𝒜 ) → 𝐻 𝑝 (𝒱 ; 𝒜 ) that commutes with all of the above maps, as you

can check, with similar homomorphisms for 𝒜 and ℬ. Thus in the limit we obtain
a sequence
(6.11)
𝛼∗ 𝛽∗ 𝛿∗
⋯ → 𝐻 𝑝 (𝑀; 𝒜 ) ⟶ 𝐻 𝑝 (𝑀; ℬ) ⟶ 𝐻𝛽 (𝑀; 𝒞 ) ⟶ 𝐻 𝑝+1 (𝑀; 𝒜 ) → ⋯ ,
𝑝

where 𝐻𝛽 (𝑀; 𝒞 ) = lim 𝐻𝛽 (𝒰; 𝒞 ) for each 𝑝. It is straightforward to check that

𝑝 𝑝
⟶
this sequence is still exact.
To complete the proof, we will show that the map

ℐ ∶ 𝐻𝛽 (𝑀; 𝒞 ) → 𝐻 𝑝 (𝑀; 𝒞 )
𝑝

induced by inclusion 𝐶𝛽 (𝒰; 𝒞 ) ↪ 𝐶 𝑝 (𝒰; 𝒞 ) is an isomorphism for each 𝑝. The

𝑝

proof will be based on the following fact, whose verification we postpone until the
end of the proof:

Given a cochain 𝑐 ∈ 𝐶 𝑝 (𝒰; 𝒞 ), there exist a refinement 𝒱 of 𝒰

and a refining map 𝜌 ∶ 𝒱 → 𝒰 such that 𝜌# 𝑐 ∈ 𝐶𝛽 (𝒱 ; 𝒞 ).
(6.12) 𝑝
The Long Exact Cohomology Sequence 157

Granting this for the moment, we prove that ℐ is bijective. To show that it
is injective, suppose ℐ ([[𝑐]]) = 0. We can choose a representative cochain 𝑐 ∈
𝐶𝛽 (𝒰; 𝒞 ) for some open cover 𝒰, and the hypothesis implies there is a refining map
𝑝

𝜌 ∶ 𝒱 → 𝒰 and a cochain 𝛾 ∈ 𝐶 𝑝−1 (𝒱 ; 𝒞 ) such that 𝜌# 𝑐 = 𝛿𝛾. By (6.12), there is

another refining map 𝜌 ̃ ∶ 𝒲 → 𝒱 such that 𝜌#̃ 𝛾 = 𝛽# 𝑏 for some 𝑏 ∈ 𝐶 𝑝−1 (𝒲 ; 𝒞 ).

𝜌#̃ 𝜌# 𝑐 = 𝜌#̃ 𝛿𝛾 = 𝛿 𝜌#̃ 𝛾 = 𝛿𝛽# 𝑏,

Then

so, after refinement, 𝑐 maps to a coboundary in 𝐶𝛽 (𝒲 ; 𝒞 ) and thus [[𝑐]] = 0 ∈

𝑝

𝐻𝛽 (𝑀; 𝒞 ).
𝑝

To show that ℐ is surjective, let [[𝑐]] ∈ 𝐻 𝑝 (𝑀; 𝒞 ) be arbitrary, represented

by a cochain 𝑐 ∈ 𝐶 𝑝 (𝒰; 𝒞 ) with 𝛿𝑐 = 0. By (6.12), there exists a refining map
𝜌 ∶ 𝒱 → 𝒰 such that 𝜌# 𝑐 = 𝛽# 𝑏 for some 𝑏 ∈ 𝐶 𝑝 (𝒱 ; ℬ). Now
𝛿𝛽# 𝑏 = 𝛿𝜌# 𝑐 = 𝜌# 𝛿𝑐 = 0,
so 𝛽# 𝑏 represents a cohomology class in 𝐻𝛽 (𝒰; 𝒞 ), and passing to the limit, we see
𝑝

that [[𝑐]] = ℐ ([[𝛽# 𝑏]]).

Thus we can replace 𝐻𝛽 (𝒰; 𝒞 ) with 𝐻 𝑝 (𝒰; 𝒞 ) in (6.11), replace 𝛽∗ by 𝛽∗ ∘ ℐ
𝑝

(which is just 𝛽∗ ∶ 𝐻 𝑝 (𝑀; ℬ) → 𝐻 𝑝 (𝑀; 𝒞 )), and replace 𝛿∗ by 𝛿∗ ∘ ℐ −1 , and we

obtain the long exact sequence (6.8).

(6.9) of sheaf morphisms with exact rows. Let 𝛾 ∈ 𝐻 𝑝 (𝑀; 𝒞 ) be arbitrary. By

Now we prove the naturality claim. Suppose we have a commutative diagram

(6.12), we can choose an open cover 𝒰 such that 𝛾 is represented by a cocycle

𝑐 ∈ 𝐶𝛽 (𝒰; 𝒞 ). By (6.6), 𝛿∗ (𝛾) is represented by a cocycle 𝑎 ∈ 𝐶 𝑝+1 (𝒰; 𝒜 ) such
𝑝

that there exists 𝑏 ∈ 𝐶 𝑝 (𝒰; ℬ) with

(6.13) 𝛽# 𝑏 = 𝑐 and 𝛼# 𝑎 = 𝛿𝑏.

Set 𝑏′ = 𝐵# 𝑏 ∈ 𝐶 𝑝 (𝒰; ℬ ′ ) and 𝑎′ = 𝐴# 𝑎 ∈ 𝐶 𝑝+1 (𝒰; 𝒜 ′ ). Note that (6.3)

shows that 𝐵# commutes with 𝛿, and commutativity of (6.9) implies that 𝛼#′ ∘ 𝐴# =
𝐵# ∘ 𝛼# and 𝛽#′ ∘ 𝐵# = 𝐶# ∘ 𝛽# . These identities together with (6.13) yield
𝛽# 𝑏′ = 𝛽#′ 𝐵# 𝑏 = 𝐶# 𝛽# 𝑏 = 𝐶# 𝑐,
𝛼#′ 𝑎′ = 𝛼#′ 𝐴# 𝑎 = 𝐵# 𝛼# 𝑎 = 𝐵# 𝛿𝑏 = 𝛿𝐵# 𝑏 = 𝛿𝑏′ ,
so it follows from (6.3) that 𝛿∗′ 𝐶∗ 𝛾 = 𝛿∗′ [[𝐶# 𝑐]] is represented by 𝑎′ . Therefore,
𝐴∗ 𝛿∗ 𝛾 = 𝐴∗ [[𝑎]] = [[𝐴# 𝑎]] = [[𝑎′ ]] = 𝛿∗′ 𝐶∗ 𝛾, which shows that diagram (6.10)
commutes as claimed.

where we use the fact that 𝑀 is a paracompact Hausdorff space.) Suppose 𝑐 ∈

It remains only to prove statement (6.12). (This is the only part of the proof

𝐶 𝑝 (𝒰; 𝒞 ) is arbitrary. After refining the cover if necessary, we can assume that
𝒰 = {𝑈𝛼 }𝛼∈𝐴 is locally finite. By [LeeTM, Lemma 4.84], for each 𝛼 ∈ 𝐴 there
exists an open set 𝑊𝛼 such that 𝑊 𝛼 ⊆ 𝑈𝛼 and the collection {𝑊𝛼 }𝛼∈𝐴 still covers
158 6. Sheaf Cohomology

𝑀. Because the cover 𝒰 is locally finite and the sheaf sequence (6.7) is exact, for
each 𝑥 ∈ 𝑀 we can choose a neighborhood 𝑉𝑥 small enough that the following
properties are satisfied:
(i) If 𝑥 ∈ 𝑈𝛼0 ∩ ⋯ ∩ 𝑈𝛼𝑝 , then 𝑉𝑥 ⊆ 𝑈𝛼0 ∩ ⋯ ∩ 𝑈𝛼𝑝 and there exists a section
𝑏𝛼0 …𝛼𝑝 ∈ ℬ(𝑉𝑥 ) such that 𝛽 (𝑏𝛼0 …𝛼𝑝 ) = 𝑐𝛼0 …𝛼𝑝 |𝑉𝑥 .
(𝑥) (𝑥)

(ii) If 𝑥 ∈ 𝑊𝛼 , then 𝑉𝑥 ⊆ 𝑊𝛼 .
(iii) If 𝑉𝑥 ∩ 𝑊𝛼 ≠ ∅, then 𝑉𝑥 ⊆ 𝑈𝛼 .
Let 𝒱 be the indexed open cover {𝑉𝑥 }𝑥∈𝑀 . By (i), it is a refinement of 𝒰.
Choose a refining map 𝜌 ∶ 𝒱 → 𝒰 such that for each 𝑥 ∈ 𝑀, we have 𝑥 ∈
𝑊𝜌(𝑥) ⊆ 𝑈𝜌(𝑥) . For any (𝑝 + 1)-tuple 𝑥0 , … , 𝑥𝑝 , define

𝑏̃𝑥0 …𝑥𝑝 = 𝑏𝜌(𝑥0 )…𝜌(𝑥 ) .

(𝑥 )
0 𝑝

Then it is straightforward to check that the cochain 𝑏 ∈ 𝐶 𝑝 (𝒱 ; ℬ) satisfies 𝛽# 𝑏 =

𝜌# 𝑐. □

Suppose 𝒮 is a sheaf of real or complex vector spaces on a topological space 𝑀.

For many sheaves, we can define an important numerical invariant as follows.

If the spaces 𝐻 𝑘 (𝑀; 𝒮 ) are all finite-dimensional and are nonzero for only finitely
many values of 𝑘, we define the Euler characteristic of 𝓢 to be the integer

𝜒(𝒮 ) = (−1)𝑘 dim 𝐻 𝑘 (𝑀; 𝒮 ).

∑
𝑘

(The name reflects the analogy with the Euler characteristic of a topological space,
which can be computed as the alternating sum of the ranks of the singular homology
groups [LeeTM, Thm. 13.36].)
Proposition 6.10. Suppose 0 → ℛ → 𝒮 → 𝒯 → 0 is a short exact sequence
of sheaves of vector spaces on a locally compact Hausdorff space 𝑀. If the Euler
characteristics 𝜒(ℛ), 𝜒(𝒮 ), and 𝜒(𝒯 ) are all defined, then

𝜒(𝒮 ) = 𝜒(ℛ) + 𝜒(𝒯 ).

Proof. By Theorem 6.9, we have a long exact sequence

⋯ → 𝐻 𝑘 (𝑀; ℛ) → 𝐻 𝑘 (𝑀; 𝒮 ) → 𝐻 𝑘 (𝑀; 𝒯 ) → 𝐻 𝑘+1 (𝑀; ℛ) → ⋯ .

For each 𝑘, let 𝑍 𝑘 (𝑀; 𝒮 ) denote the kernel of the homomorphism 𝐻 𝑘 (𝑀; 𝒮 ) →
𝐻 𝑘 (𝑀; 𝒯 ), which is also the image of 𝐻 𝑘 (𝑀; ℛ) → 𝐻 𝑘 (𝑀; 𝒮 ); and define
𝑍 𝑘 (𝑀; ℛ) and 𝑍 𝑘 (𝑀; 𝒯 ) similarly. It follows from the rank-nullity law of linear
algebra (which says that for a linear map 𝐹 ∶ 𝑉 → 𝑊 between finite-dimensional
vector spaces, the dimension of 𝑉 is equal to the sum of the dimensions of the
Acyclic Resolutions 159

kernel and the image of 𝐹 ) that

dim 𝐻 𝑘 (𝑀; ℛ) = dim 𝑍 𝑘 (𝑀; ℛ) + dim 𝑍 𝑘 (𝑀; 𝒮 ),
dim 𝐻 𝑘 (𝑀; 𝒮 ) = dim 𝑍 𝑘 (𝑀; 𝒮 ) + dim 𝑍 𝑘 (𝑀; 𝒯 ),
dim 𝐻 𝑘 (𝑀; 𝒯 ) = dim 𝑍 𝑘 (𝑀; 𝒯 ) + dim 𝑍 𝑘+1 (𝑀; ℛ).

Multiplying each of these equations by (−1)𝑘 and summing over 𝑘 gives formulas
for the respective Euler characteristics. When we add the first and last of these

cancel, yielding 𝜒(ℛ) − 𝜒(𝒮 ) + 𝜒(𝒯 ) = 0.

equations and subtract the second, we find that all the terms on the right-hand side
□

Acyclic Resolutions
The definition of the sheaf cohomology groups is too abstract to be useful for com-
putations in most circumstances. In this section, we introduce an important tool

Suppose 𝒮 is a sheaf of abelian groups on a topological space 𝑀. A resolution

that often makes computations straightforward.

of 𝓢 is a (possibly infinite) exact sequence of sheaves of abelian groups and sheaf

morphisms of the form
(6.14) 0 → 𝒮 → 𝒜0 → 𝒜1 → 𝒜2 → ⋯ .
If 𝒮 is a sheaf of real or complex vector spaces, then we require sheaves and mor-

is a resolution of the constant sheaf ℂ.

phisms in the same category. Thus, for example, the de Rham sheaf sequence (5.17)

A sheaf 𝒜 of abelian groups on 𝑀 is said to be acyclic if 𝐻 𝑞 (𝑀; 𝒜 ) = 0 for

all 𝑞 ≥ 1. (Of course, 𝐻 0 (𝑀; 𝒜 ) ≅ 𝒜 (𝑀), so we cannot expect that to be zero

sheaves 𝒜 𝑘 is acyclic for 𝑘 ≥ 0.

in most cases.) A resolution (6.14) is called an acyclic resolution if each of the

The main tool for computing sheaf cohomology groups is the following theo-
rem. It is named the de Rham–Weil theorem because André Weil [Wei52] intro-
duced this technique as a way of proving the de Rham theorem (see Theorem 6.20
below).
Theorem 6.11 (De Rham–Weil). Suppose 𝒮 is a sheaf of abelian groups on a
paracompact Hausdorff space 𝑀 and
𝜄 𝑑 𝑑
(6.15) 0 → 𝒮 ⟶ 𝒜0 ⟶ 𝒜1 ⟶ 𝒜2 → ⋯
is an acyclic resolution of 𝒮 . Then the sequence of global sections
𝑑 𝑑
0 → 𝒜 0 (𝑀) ⟶ 𝒜 1 (𝑀) ⟶ 𝒜 2 (𝑀) → ⋯
160 6. Sheaf Cohomology

is a cochain complex, and for each 𝑞, the sheaf cohomology group 𝐻 𝑞 (𝑀; 𝒮 ) is
isomorphic to the cohomology group 𝐻 𝑞 (𝒜 ∗ (𝑀)) of this complex; more precisely,

𝐻 0 (𝑀; 𝒮 ) ≅ Ker (𝑑 ∶ 𝒜 0 (𝑀) → 𝒜 1 (𝑀)), and

Ker (𝑑 ∶ 𝒜 (𝑀) → 𝒜
𝑞 𝑞+1
(𝑀))
𝐻 𝑞 (𝑀; 𝒮 ) ≅ for 𝑞 ≥ 1.
Im (𝑑 ∶ 𝒜 𝑞−1 (𝑀) → 𝒜 𝑞 (𝑀))
If (6.15) is a resolution of sheaves of real or complex vector spaces, then the iso-

Suppose 𝐹 ∶ 𝒮 → 𝒯 is a morphism between sheaves of abelian groups on 𝑀 and

morphisms are linear. The isomorphisms satisfy the following naturality property:

we have a commutative diagram of sheaf homomorphisms in which both horizontal

rows are acyclic resolutions:

𝜄 / 0 𝑑 / 1 𝑑 / 2 𝑑 /
0 /𝒮 𝒜 𝒜 𝒜 ⋯
(6.16) 𝐹 𝜑0  𝜑1  𝜑2 
0 /𝒯 / ℬ0 / ℬ1 / ℬ2 / ⋯.
𝜄 ′
𝑑 ′
𝑑 ′
𝑑′


Then for each 𝑞 ≥ 0, the following diagram commutes:

≅ / 𝑞 ∗
𝐻 𝑞 (𝑀; 𝒮 ) 𝐻 (𝒜 (𝑀))
𝐹∗ 𝜑∗
𝑞

≅ / 𝑞 ∗
(6.17)
𝐻 𝑞 (𝑀; 𝒯 ) 𝐻 (ℬ (𝑀)),
 

where 𝜑∗ is the cohomology homomorphism induced by the global section homo-

𝑞

morphisms 𝜑𝑞 ∶ 𝒜 𝑞 (𝑀) → ℬ 𝑞 (𝑀).

is a cochain complex follows from Lemma 5.20, and the claim about 𝐻 0 (𝑀; 𝒮 )
Proof. Given an acyclic resolution (6.15), the fact that the global section sequence

follows from Proposition 5.24. From now on, we assume 𝑞 ≥ 1.

For each 𝑘 ≥ 0, define a sheaf 𝒵 𝑘 by

𝒵 𝑘 = Ker (𝑑 ∶ 𝒜 𝑘 → 𝒜 𝑘+1 ).

This is a subsheaf of 𝒜 𝑘 by Exercise 5.17, but not necessarily acyclic. In particular,

𝒵 0 is isomorphic to 𝒮 by Proposition 5.22.
For each 𝑘 ≥ 0, we get a short exact sequence
𝑑
0 → 𝒵 𝑘 ↪ 𝒜 𝑘 ⟶ 𝒵 𝑘+1 → 0.

The associated long exact sequence reads in part

𝐻 𝑞−1 (𝑀; 𝒜 𝑘 ) → 𝐻 𝑞−1 (𝑀; 𝒵 𝑘+1 ) → 𝐻 𝑞 (𝑀; 𝒵 𝑘 ) → 𝐻 𝑞 (𝑀; 𝒜 𝑘 ).

Acyclic Resolutions 161

When 𝑞 > 1, the groups on both ends are zero because 𝒜 𝑘 is acyclic; therefore we
have isomorphisms

(6.18) 𝐻 𝑞−1 (𝑀; 𝒵 𝑘+1 ) ≅ 𝐻 𝑞 (𝑀; 𝒵 𝑘 ) for 𝑞 > 1.

To prove the theorem, we proceed as follows. Assuming 𝑞 > 1, we apply (6.18)

repeatedly to conclude

𝐻 𝑞 (𝑀; 𝒮 ) ≅ 𝐻 𝑞 (𝑀; 𝒵 0 )
≅ 𝐻 𝑞−1 (𝑀; 𝒵 1 )
≅ 𝐻 𝑞−2 (𝑀; 𝒵 2 )
⋮
≅ 𝐻 1 (𝑀; 𝒵 𝑞−1 ),

and the same conclusion holds trivially when 𝑞 = 1. At this point, we can no longer
use (6.18). Instead, we have the exact sequence
𝑑 𝛿∗
𝐻 0 (𝑀; 𝒜 𝑞−1 ) ⟶ 𝐻 0 (𝑀; 𝒵 𝑞 ) ⟶ 𝐻 1 (𝑀; 𝒵 𝑞−1 ) → 𝐻 1 (𝑀; 𝒜 𝑞−1 ).

Here the right-hand group is zero, which means that 𝛿∗ is surjective, and exactness
implies

𝐻 0 (𝑀; 𝒵 𝑞 ) 𝒵 𝑞 (𝑀) Ker (𝑑 ∶ 𝒜 (𝑀) → 𝒜 (𝑀))

𝑞 𝑞+1
𝐻 1 (𝑀; 𝒵 𝑞−1 ) ≅ ≅ ≅ ,
Ker 𝛿∗ Im 𝑑 Im (𝑑 ∶ 𝒜 𝑞−1 (𝑀) → 𝒜 𝑞 (𝑀))
which is what we needed to prove. In the case of sheaves of vector spaces, all the
maps above are linear.

form (6.16). For each 𝑘, let 𝒲 𝑘 ⊆ ℬ 𝑘 be the kernel of 𝑑 ′ ∶ ℬ 𝑘 → ℬ 𝑘+1 . Since

To prove the naturality claim, suppose we have a commutative diagram of the

(6.16) commutes, 𝜑𝑘 maps 𝒵 𝑘 to 𝒲 𝑘 for each 𝑘, and we have commutative dia-

grams of sheaf morphisms:

0 / 𝒵𝑘  / 𝒜𝑘 / 𝒵 𝑘+1 /0

(6.19) 𝜑𝑘 𝜑𝑘 𝜑𝑘+1
0 / 𝒲 𝑘  / ℬ𝑘 / 𝒲 𝑘+1 / 0,
  

propriate subsheaves. Because the isomorphisms 𝐻 𝑞 (𝑀; 𝒮 ) ≅ 𝐻 𝑞 (𝒜 ∗ (𝑀)) and

in which the first and third vertical maps are understood to be restricted to the ap-

𝐻 𝑞 (𝑀; 𝒯 ) ≅ 𝐻 𝑞 (ℬ ∗ (𝑀)) are obtained as compositions of (restrictions of) con-

necting homomorphisms in the long exact sequences associated with the rows of
(6.19), the commutativity of (6.17) follows by repeatedly applying the naturality
result of Theorem 6.9. □
162 6. Sheaf Cohomology

Fine Sheaves
To apply the de Rham–Weil theorem, we need to have a good supply of acyclic
sheaves. We start with an important example.
Example 6.12 (Sheaves of Smooth Forms Are Acyclic). Let 𝑀 be a smooth
manifold. We will show that each sheaf ℰ 𝑘 of smooth complex-valued 𝑘-forms

To prove this claim, it suffices to show that 𝐻 𝑝 (𝒰; ℰ 𝑘 ) = 0 for every open
is acyclic.

cover 𝒰 of 𝑀. Let 𝒰 = {𝑈𝛼 }𝛼∈𝐴 be an arbitrary open cover, and choose a

smooth partition of unity {𝜑𝛼 }𝛼∈𝐴 subordinate to 𝒰. Define a homomorphism
𝜃 ∶ 𝐶 𝑝 (𝒰; ℰ 𝑘 ) → 𝐶 𝑝−1 (𝒰; ℰ 𝑘 ) by

(𝜃𝑐)𝛼0 …𝛼𝑝−1 = 𝜑𝛽 𝑐𝛽𝛼0 …𝛼𝑝−1 ,

∑
(6.20)
𝛽∈𝐴

where the 𝑘-form 𝜑𝛽 𝑐𝛽𝛼0 …𝛼𝑝−1 is extended to all of 𝑈𝛼0 ∩ ⋯ ∩ 𝑈𝛼𝑝−1 by defining it
to be zero outside the support of 𝜑𝛽 . Because this is a finite sum of smooth 𝑘-forms
in a neighborhood of each point, it defines a smooth form on 𝑈𝛼0 ∩ ⋯ ∩ 𝑈𝛼𝑝−1 and
thus a (𝑝 − 1)-cochain. We will show that 𝜃 satisfies a cochain homotopy formula
𝛿𝜃 + 𝜃𝛿 = Id𝐶 𝑝 (𝒰;ℰ 𝑘 ) , from which it follows that every cocycle in 𝐶 𝑝 (𝒰; ℰ 𝑘 ) is a
coboundary, thus proving the claim.
To verify the cochain homotopy formula, we compute
𝑝
(𝛿𝜃𝑐)𝛼0 …𝛼𝑝 = (−1)𝑗 (𝜃𝑐)𝛼0 …𝛼̂𝑗 …𝛼𝑝
∑
𝑗=0
𝑝
= (−1)𝑗 𝜑𝛽 𝑐𝛽𝛼0 …𝛼̂𝑗 …𝛼𝑝 ,
∑∑
𝑗=0 𝛽

(𝜃𝛿𝑐)𝛼0 …𝛼𝑝 = 𝜑𝛽 (𝛿𝑐)𝛽𝛼0 …𝛼𝑝

∑
𝛽
𝑝
= 𝜑𝛽 𝑐𝛽𝛼̂ 0 …𝛼𝑝 + (−1)𝑗+1 𝜑𝛽 𝑐𝛽𝛼0 …𝛼̂𝑗 …𝛼𝑝 .
∑ ∑∑
𝛽 𝛽 𝑗=0

in the last line, which is equal to 𝑐𝛼0 …𝛼𝑝 because ∑𝛽 𝜑𝛽 ≡ 1.

When these expressions are added together, everything cancels except the first sum
//
The same argument generalizes in an obvious way to any sheaf of smooth sec-
tions of a smooth vector bundle on a smooth manifold, and to any sheaf of con-
tinuous sections of a topological vector bundle on a paracompact Hausdorff space
(which is the type of space on which partitions of unity can be constructed; see
[LeeTM, Thm. 4.85]). Since the purpose of sheaf cohomology is to determine
when local objects can be patched together to form global ones, you should think
Acyclic Resolutions 163

of these phenomena as expressions of the fact that there are no obstructions to patch-

The preceding example can be generalized significantly. Given a sheaf 𝒮 of

ing things together when partitions of unity are available.

abelian groups on a topological space 𝑀 and a locally finite indexed open cover
𝒰 = {𝑈𝛽 }𝛽∈𝐵 , we define a sheaf partition of unity subordinate to 𝓤 to be an
indexed collection of sheaf morphisms 𝜂𝛽 ∶ 𝒮 → 𝒮 satisfying the following two
conditions:
(i) For each 𝛽 ∈ 𝐵, the support of 𝜂𝛽 is contained in 𝑈𝛽 (where the support
of a sheaf morphism 𝐹 ∶ 𝒮 → 𝒯 is the closure of the set of points 𝑥 ∈ 𝑀
such that the stalk homomorphism 𝐹𝑥 is nonzero).
(ii) For each 𝑥, ∑𝛽∈𝐵 (𝜂𝛽 )𝑥 = Id𝒮𝑥 .

hood on which 𝜂𝛽 = 0 for all but finitely many 𝛽, so the sum in (ii) has only finitely
(The fact that the open cover is locally finite ensures that each point has a neighbor-

many nonzero terms.) A sheaf 𝒮 is said to be fine if for every locally finite open
cover there exists a sheaf partition of unity subordinate to it. To avoid confusion,

continuous functions taking values in [0, 1] whose sum is 1 and whose supports are
we will refer to a partition of unity in the usual topological sense (a collection of

locally finite) as a topological partition of unity; if the underlying space is a smooth

manifold and each of the functions is smooth, it will be called a smooth partition
of unity.
Example 6.13 (Fine Sheaves).
(a) On a smooth manifold 𝑀, the sheaf ℰ of smooth complex-valued func-
tions is fine: given a locally finite open cover 𝒰 = {𝑈𝛽 }𝛽∈𝐵 , let {𝜓𝛽 }𝛽∈𝐵
be a subordinate smooth partition of unity, and define 𝜂𝛽 ∶ ℰ → ℰ by
sending 𝑓 ∈ ℰ (𝑈 ) to 𝜓𝛽 𝑓 . By the same argument, every sheaf of ℰ -
modules on 𝑀 is fine, as is every sheaf of modules over the sheaf ℰℝ of
real-valued smooth functions. The most important examples are sheaves

(b) If 𝑀 is a paracompact Hausdorff space, the sheaf 𝒞 of continuous com-

of sections of smooth vector bundles.

sheaf of 𝒞 -modules on 𝑀.
plex-valued functions is fine by essentially the same argument, as is any

(c) Every skyscraper sheaf is fine. Let 𝐺𝑝 be a skyscraper sheaf on a topo-

logical space 𝑀 supported at 𝑝 ∈ 𝑀. Given a locally finite open cover
{𝑈𝛼 }𝛼∈𝐴 of 𝑀, choose one index 𝛽 ∈ 𝐴 such that 𝑝 ∈ 𝑈𝛽 , and define 𝜂𝛽
to be the identity morphism and 𝜂𝛼 = 0 for all 𝛼 ≠ 𝛽. //
On the other hand, constant sheaves and sheaves of 𝒪-modules are almost never
fine (see Problem 6-1).
Proposition 6.14. If 𝒮 is a fine sheaf on a paracompact Hausdorff space 𝑀, then
𝒮 is acyclic.
164 6. Sheaf Cohomology

Proof. Let 𝒮 be such a sheaf. Since every open cover of 𝑀 has a locally finite
open refinement, to prove the proposition it suffices to show that 𝐻 𝑘 (𝒰; 𝒮 ) = 0
for every 𝑘 ≥ 1 and every locally finite open cover 𝒰. Given a locally finite open
cover 𝒰 = {𝑈𝛽 }𝛽∈𝐵 , let {𝜂𝛽 }𝛽∈𝐵 be a sheaf partition of unity subordinate to it. For
each of the sheaf morphisms 𝜂𝛽 and any open subset 𝑉 ⊆ 𝑀, the homomorphism
𝜂𝛽 ∶ 𝒮 (𝑉 ∩𝑈𝛽 ) → 𝒮 (𝑉 ∩𝑈𝛽 ) extends to a homomorphism 𝜂 𝛽̃ ∶ 𝒮 (𝑉 ∩𝑈𝛽 ) → 𝒮 (𝑉 )
by requiring that
𝜂 𝛽̃ (𝑐)|𝑉 ∩𝑈 = 𝜂𝛽 (𝑐) and 𝜂 𝛽̃ (𝑐)|𝑉 ∖supp 𝜂 = 0;
𝛽 𝛽

For each 𝑝 ≥ 1, define a map 𝜃 ∶ 𝐶 𝑝 (𝒰; 𝒮 ) → 𝐶 𝑝−1 (𝒰; 𝒮 ) by analogy with

the gluing and locality properties of sheaves guarantee that this is uniquely defined.

(𝜃𝑐)𝛼0 …𝛼𝑝−1 = 𝜂̃ 𝑐 ,
(6.20):
∑ 𝛽 ( 𝛽𝛼0 …𝛼𝑝−1 )
𝛽∈𝐵
where we interpret the sum on the right-hand side by noting that 𝑈𝛼0 ∩ ⋯ ∩ 𝑈𝛼𝑝−1
has an open cover such that only finitely many terms of this sum are nonzero on

patch together to determine an element of 𝒮 (𝑈𝛼0 ∩ ⋯ ∩ 𝑈𝛼𝑝−1 ). The rest of the proof
each open set of the cover, and the gluing property ensures that these finite sums

of the proposition is exactly like the argument in Example 6.12. □

of ℰ -modules and skyscraper sheaves.

In complex manifold theory, the acyclic sheaves that will concern us are sheaves

Sheaf Cohomology and Singular Cohomology

For the next result about sheaf cohomology, we need a few more definitions. We
begin with a very brief review of singular homology and cohomology theory.
For much more detail on the subject, consult any algebraic topology text such as
[Hat02] or [Mun84].

Singular Homology
For a nonnegative integer 𝑘, the standard 𝒌-simplex Δ𝑘 is the convex hull of
the 𝑘 + 1 points {𝑒0 , 𝑒1 , … , 𝑒𝑘 } in ℝ𝑘 , where 𝑒0 = 0 and 𝑒𝑗 is the 𝑗th standard basis
vector for 𝑗 ≥ 1. Since every point in the convex hull can be written as a linear
combination ∑𝑘𝑗=0 𝑡𝑗 𝑒𝑗 with ∑𝑘𝑗=0 𝑡𝑗 = 1, we can express Δ𝑘 more explicitly as
𝑘
Δ𝑘 = {(𝑡1 , … , 𝑡𝑘 ) ∈ ℝ𝑘 ∶ 0 ≤ 𝑡𝑗 ≤ 1 and 𝑡 𝑗 ≤ 1 }.
∑
𝑗=1

Let 𝑀 be a topological space. A singular 𝒌-simplex in 𝑀 is a continuous map

𝜎 ∶ Δ𝑘 → 𝑀. The free abelian group on the set of all singular 𝑘-simplices is called
Sheaf Cohomology and Singular Cohomology 165

the singular chain group in dimension 𝑘; we will denote it by Sing𝑘 (𝑀). An ele-
ment of that group, called a singular 𝒌-chain, is a finite formal linear combination
of singular 𝑘-simplices with integer coefficients.
The boundary of a singular 𝑘-simplex 𝜎 is the singular (𝑘−1)-chain 𝜕𝜎 defined
by
𝑘
𝜕𝜎 = (−1)𝑖 𝜎 ∘ 𝐹𝑖,𝑘 ,
∑
𝑖=0
where 𝐹𝑖,𝑘 ∶ Δ𝑘−1 → Δ𝑘 , called the 𝑖th face map in dimension 𝑘, is the restriction
of the unique affine map from ℝ𝑘−1 to ℝ𝑘 that sends the vertices 𝑒0 , … , 𝑒𝑘−1 to
𝑒0 , … , 𝑒̂𝑖 , … , 𝑒𝑘 , respectively, so it maps Δ𝑘−1 homeomorphically onto the face of
Δ𝑘 opposite 𝑒𝑖 . The boundary operator extends by linearity to a group homomor-
phism 𝜕 ∶ Sing𝑘 (𝑀) → Sing𝑘−1 (𝑀). A computation shows that 𝜕 ∘ 𝜕 = 0, and
therefore it makes sense to define the 𝒌th singular homology group of 𝑴 by
Ker (𝜕 ∶ Sing𝑘 (𝑀) → Sing𝑘−1 (𝑀))
𝐻𝑘 (𝑀) = .
Im (𝜕 ∶ Sing𝑘+1 (𝑀) → Sing𝑘 (𝑀))

A singular 𝑘-chain 𝑐 is called a cycle if it satisfies 𝜕𝑐 = 0, and a boundary if

𝑐 = 𝜕𝑏 for some (𝑘 − 1)-chain 𝑏. Every cycle 𝑐 determines an element of 𝐻𝑘 (𝑀),
called its homology class and denoted by [𝑐]; any other cycle differing from 𝑐 by a
boundary represents the same homology class. Two cycles that differ by a boundary
are said to be homologous.

pose 𝑀 is a compact, connected, orientable smooth 𝑛-manifold. Every such mani-

We need the following basic fact about singular homology of manifolds. Sup-

fold admits a smooth triangulation, which is a singular 𝑛-cycle 𝑐 = ∑𝑗 𝜎𝑗 with the

property that each 𝜎𝑗 ∶ Δ𝑛 → 𝑀 is a smooth orientation-preserving embedding,
and the images of two distinct 𝜎𝑗 ’s intersect only along their boundaries if at all.
(See [Mun66, Thm. 10.6] for a proof.) The cohomology class of 𝑐 depends only on
the orientation of 𝑀; it is denoted by [𝑀] and called the fundamental class of 𝑴.
For a proof of the following proposition, see [Mun84, Cor. 65.3] or [Hat02, Thm.
3.26]).
Proposition 6.15. Suppose 𝑀 is a compact, connected, oriented smooth 𝑛-dimen-
sional manifold, and [𝑀] is its fundamental class. Then 𝐻𝑛 (𝑀) is an infinite cyclic
group generated by [𝑀].

Singular Cohomology
If 𝐺 is an abelian group, a singular 𝒌-cochain in 𝑴 with coefficients in 𝑮
is a group homomorphism 𝜑 ∶ Sing𝑘 (𝑀) → 𝐺. The set of all such cochains, de-
noted by Sing𝑘 (𝑀; 𝐺) = Hom ( Sing𝑘 (𝑀), 𝐺), is a group under pointwise addi-
tion: (𝜑 + 𝜑′ )(𝑐) = 𝜑(𝑐) + 𝜑′ (𝑐). If in addition 𝐺 is a real or complex vector
166 6. Sheaf Cohomology

space, then Sing𝑘 (𝑀; 𝐺) is a vector space under pointwise scalar multiplication. A

The dual of the boundary map is a group homomorphism 𝛿 ∶ Sing𝑘 (𝑀; 𝐺) →

singular cochain is uniquely determined by its action on each singular simplex.

Sing𝑘+1 (𝑀; 𝐺) called the coboundary operator; it is defined by

(𝛿𝜑)(𝑐) = 𝜑(𝜕𝑐).

A singular cochain 𝜑 satisfying 𝛿𝜑 = 0 is called a (singular) cocycle, and one

satisfying 𝜑 = 𝛿𝜓 for some cochain 𝜓 is a (singular) coboundary. The fact that
𝜕 ∘ 𝜕 = 0 implies immediately that 𝛿 ∘ 𝛿 = 0, so every coboundary is a cocycle.
Thus we can form the following quotient space, called the 𝒌th singular cohomology
group of 𝑴 with coefficients in 𝑮:

Ker (𝛿 ∶ Sing𝑘 (𝑀; 𝐺) → Sing𝑘+1 (𝑀; 𝐺))

𝐻Sing
𝑘
(𝑀; 𝐺) = .
Im (𝛿 ∶ Sing𝑘−1 (𝑀; 𝐺) → Sing𝑘 (𝑀; 𝐺))

If 𝐺 is a real or complex vector space, then so is 𝐻Sing

𝑘
(𝑀; 𝐺).
Every homomorphism 𝐹 ∶ 𝐺 → 𝐻 between abelian groups induces a homo-
morphism 𝐹# ∶ Sing𝑘 (𝑀; 𝐺) → Sing𝑘 (𝑀; 𝐻) by 𝐹# (𝜑)(𝑐) = 𝐹 (𝜑(𝑐)) for each
singular cochain 𝜑 and singular chain 𝑐. This homomorphism commutes with the

phism 𝐹∗ ∶ 𝐻Sing
𝑘
(𝑀; 𝐺) → 𝐻Sing 𝑘
(𝑀; 𝐻), called a coefficient homomorphism.
coboundary operators, so it defines a cochain map and thus descends to a homomor-

For a fixed space 𝑀, the coefficient homomorphisms satisfy

(Id𝐺 )∗ = Id𝐻 𝑘 (𝑀;𝐺) and (𝐹 ∘ 𝐹 ′ )∗ = 𝐹∗ ∘ 𝐹∗′ ,

Sing

so the assignment 𝐺 ↦ 𝐻Sing 𝑘

(𝑀; 𝐺), 𝐹 ↦ 𝐹∗ is a covariant functor from the
category of abelian groups to itself, with a similar statement for the category of real
or complex vector spaces.
It is a standard result in algebraic topology that the singular homology and

fact is that if 𝑀 is a contractible space, then its singular cohomology groups agree
cohomology groups are homotopy invariants. One important consequence of this

with those of a one-point space, namely 𝐻Sing 0

(𝑀; 𝐺) ≅ 𝐺 and 𝐻Sing
𝑘
(𝑀; 𝐺) = 0
for 𝑘 > 0.

map from Sing𝑘 (𝑀; 𝐺) × Sing𝑘 (𝑀) to 𝐺, carrying (𝜑, 𝑐) to 𝜑(𝑐). This descends to
The natural action of singular cochains on singular chains defines a bilinear

a bilinear map
𝐻Sing
𝑘
(𝑀; 𝐺) × 𝐻𝑘 (𝑀) → 𝐺,

called the Kronecker pairing and denoted by ⟨[𝜑], [𝑐]⟩ = 𝜑(𝑐). To see that this is
well-defined, note that because 𝑐 is a cycle, if 𝜑 = 𝛿𝜓 is a coboundary we have
Sheaf Cohomology and Singular Cohomology 167

⟨[𝜑], [𝑐]⟩ = ⟨[𝛿𝜓], [𝑐]⟩ = (𝛿𝜓)(𝑐) = 𝜓(𝜕𝑐) = 0; and similarly if 𝑐 is a boundary

then ⟨[𝜑], [𝑐]⟩ = 0. Using this pairing, we define the Kronecker homomorphism

𝜅 ∶ 𝐻Sing
𝑘
(𝑀; 𝐺) → Hom(𝐻𝑘 (𝑀), 𝐺)

by
𝜅([𝜑])([𝑐]) = ⟨[𝜑], [𝑐]⟩.

The singular cohomology groups of a space contain exactly the same informa-
tion as the homology groups, but arranged in a different way. The precise state-
ment of this fact, called the universal coefficient theorem ([Hat02, Chap. 3] or
[Mun84, §53]), gives explicit formulas for the cohomology groups in terms of the
homology groups. We do not need the full strength of that theorem, but we will
need the following two consequences.
Proposition 6.16 (Universal Coefficient Theorem, Special Case). Let 𝑀 be
a topological space and 𝐺 be an abelian group, and let 𝜅 ∶ 𝐻Sing
𝑘
(𝑀; 𝐺) →
Hom(𝐻𝑘 (𝑀); 𝐺) be the Kronecker homomorphism in degree 𝑘.
(a) 𝜅 is surjective.
(b) If 𝐺 is a field, then 𝜅 is a vector space isomorphism.
(c) If 𝐻𝑘−1 (𝑀) is a free abelian group, then 𝜅 is a group isomorphism.
The following consequence of the universal coefficient theorem is proved in
[Hat02, Cor. 3.4] or [Mun84, Thm. 45.5].

Suppose 𝐴∗ and 𝐵∗ are chain complexes of free abelian groups and 𝜑 ∶ 𝐴∗ → 𝐵∗

Proposition 6.17 (Homology Isomorphisms Yield Cohomology Isomorphisms).

is a chain map, meaning that the following diagram commutes:

𝜕 / 𝜕 𝜕 / 𝜕
⋯ 𝐴𝑘+1 /𝐴
𝑘 𝐴𝑘−1 /⋯

𝜑 𝜑 𝜑
⋯ /𝐵 /𝐵 /𝐵 / ⋯.
𝜕 𝑘+1
𝜕 𝑘
𝜕 𝑘−1
𝜕
  

Let 𝐺 be an abelian group or a real or complex vector space, and define 𝐴𝑘 =

Hom(𝐴𝑘 , 𝐺) and 𝐵 𝑘 = Hom(𝐵𝑘 , 𝐺). Consider the diagram

𝛿 / 𝑘−1 𝛿 / 𝑘 𝛿 / 𝑘+1 𝛿 /
⋯ 𝐴O 𝐴O 𝐴O ⋯
𝜑# 𝜑# 𝜑#
⋯ / 𝐵 𝑘−1 / 𝐵𝑘 / 𝐵 𝑘+1 / ⋯,
𝛿 𝛿 𝛿 𝛿
168 6. Sheaf Cohomology

where (𝜑# 𝛾)(𝑐) = 𝛾(𝜑𝑐), and (𝛿𝛾)(𝑐) = 𝛾(𝜕𝑐). If the induced homology homo-
morphisms 𝜑∗ ∶ 𝐻𝑘 (𝐴∗ ) → 𝐻𝑘 (𝐵∗ ) are all isomorphisms, then so are the induced
cohomology homomorphisms 𝜑∗ ∶ 𝐻 𝑘 (𝐵 ∗ ) → 𝐻 𝑘 (𝐴∗ ).

Cohomology of Constant Sheaves

The next theorem shows that for a class of spaces that includes all manifolds, the
sheaf cohomology groups of a constant sheaf agree with the singular cohomology
groups. We need one last definition before introducing the theorem: a topological
space is said to be locally contractible if it has a basis of contractible open subsets.
For example, every topological manifold is locally contractible.

Theorem 6.18 (Cohomology of Constant Sheaves). Suppose 𝑀 is a locally con-

tractible paracompact Hausdorff space, 𝐺 is an abelian group, and 𝐺 is the cor-
responding constant sheaf on 𝑀. Then for each 𝑘 ≥ 0, 𝐻 𝑘 (𝑀; 𝐺) is isomorphic
to 𝐻Sing
𝑘
(𝑀; 𝐺). If 𝐺 is a real or complex vector space, then the isomorphism is
linear. The isomorphisms are natural in the following sense: if 𝐹 ∶ 𝐺 → 𝐻 is a

tor spaces) and 𝐹 ∶ 𝐺 → 𝐻 is the corresponding morphism between the constant

homomorphism of abelian groups (or a linear map between real or complex vec-

sheaves as in Example 5.5(d), then the following diagram commutes:

≅ / 𝑘
𝐻 𝑘 (𝑀; 𝐺) 𝐻Sing (𝑀; 𝐺)
𝐹∗ 𝐹∗
≅ / 𝑘
(6.21)
𝐻 𝑘 (𝑀; 𝐻) 𝐻Sing (𝑀; 𝐻).
 

Proof. Let 𝐺 be fixed. For each 𝑘 ≥ 0, we define a presheaf Sing𝑘 on 𝑀 by 𝑈 ↦

Sing𝑘 (𝑈 ; 𝐺), the group of singular 𝑘-cochains on 𝑈 , with restriction maps 𝑟𝑈
𝑉 given
by restricting cochains to singular chains in 𝑉 . Let Sing𝑘,+ be its sheafification.
Because the singular coboundary operator 𝛿 commutes with restrictions, it induces
a presheaf morphism 𝛿 ∶ Sing𝑘 → Sing𝑘+1 for each 𝑘, and corresponding sheaf
morphisms 𝛿 + ∶ Sing𝑘,+ → Sing𝑘+1,+ .
Consider the following sequence of sheaves on 𝑀:

𝜄 𝛿+ 𝛿+
(6.22) 0 → 𝐺 ⟶ Sing0,+ ⟶ Sing1,+ ⟶ ⋯ ,

where 𝜄 is obtained from the presheaf morphism that maps a locally constant func-
tion 𝑓 ∶ 𝑈 → 𝐺 to the 0-cochain that assigns the value 𝑓 (𝜎(0)) to each singular
0-simplex 𝜎 ∶ Δ0 → 𝑈 . We will show that this sequence is an acyclic resolution
of 𝐺.
Sheaf Cohomology and Singular Cohomology 169

Exactness of the sequence at 𝐺 (i.e., injectivity of 𝜄) is immediate. To prove

exactness at Sing0,+ , note that a singular 0-chain 𝑐 ∈ Sing0 (𝑈 ; 𝐺) is an arbitrary
function from points in 𝑈 to 𝐺, and it satisfies 𝛿𝑐 = 0 if and only if 𝑐(𝑥) = 𝑐(𝑦) for
all 𝑥 and 𝑦 in the same path component of 𝑈 . Since we are assuming 𝑀 is locally

𝑐 is locally constant. This means that a cochain 𝜑 ∈ Sing0 (𝑈 ; 𝐺) satisfies 𝛿𝜑 = 0 if

contractible, it is in particular locally path-connected, so this is the same as saying

and only if it is locally constant, so exactness already holds at Sing0 on the presheaf

To see that the sequence is exact at Sing𝑘,+ for 𝑘 ≥ 1, observe first that 𝛿 + ∘𝛿 + =
level.

0 because 𝛿 ∘ 𝛿 = 0. Suppose Φ ∈ Sing𝑘,+ (𝑈 ; 𝐺) satisfies 𝛿 + Φ = 0. For each

𝑥 ∈ 𝑈 , Φ(𝑥) is the germ of a 𝑘-cochain 𝜑 on a neighborhood 𝑉 of 𝑥 satisfying
𝛿𝜑 = 0. Since 𝑀 is locally contractible, we can shrink 𝑉 if necessary so that it
is contractible. Since 𝐻Sing
𝑘
(𝑉 ; 𝐺) = 0, it follows that 𝜑 is equal to 𝛿𝛽 for some
𝛽 ∈ Sing𝑘−1 (𝑉 ; 𝐺); so the sheaf sequence is exact by Lemma 5.20.
To see that the sheaves Sing𝑘,+ are acyclic, we will show they are fine. Let
𝒰 = {𝑈𝛼 }𝛼∈𝐴 be any locally finite open cover of 𝑀, and let {𝜓𝛼 } be a sub-

of unity {Ψ𝛼 } subordinate to 𝒰 by choosing a well-ordering of the index set 𝐴

ordinate topological partition of unity. Define another (discontinuous) partition

and letting Ψ𝛼 (𝑥) = 1 if 𝛼 is the least index for which 𝜓𝛼 (𝑥) > 0, and otherwise
Ψ𝛼 (𝑥) = 0. Note that the support of Ψ𝛼 is contained in that of 𝜓𝛼 and thus in 𝑈𝛼 , and
∑𝛼 Ψ𝛼 (𝑥) = 1 for all 𝑥 because exactly one term in the sum is equal to 1 and the rest
are zero. Then we define a collection of presheaf morphisms 𝜂𝛼 ∶ Sing𝑘 → Sing𝑘
by

𝜂𝛼 (𝜑)(𝜎) = Ψ𝛼 (𝜎(0))𝜑(𝜎)

for any 𝜑 ∈ Sing𝑘 (𝑈 ; 𝐺) and any singular 𝑘-simplex 𝜎 ∶ Δ𝑘 → 𝑈 ; here 𝜎(0) is the
image of 0 ∈ Δ𝑘 under the map 𝜎. The associated sheaf morphisms 𝜂𝛼+ ∶ Sing𝑘,+ →
Sing𝑘,+ form a sheaf partition of unity subordinate to 𝒰, so Sing𝑘,+ is fine.
It follows from the de Rham–Weil theorem that for each 𝑘 ≥ 1,

Ker (𝛿 + ∶ Sing𝑘,+ (𝑀; 𝐺) → Sing𝑘+1,+ (𝑀; 𝐺))

𝐻 𝑘 (𝑀; 𝐺) ≅ .
Im (𝛿 + ∶ Sing𝑘−1,+ (𝑀; 𝐺) → Sing𝑘,+ (𝑀; 𝐺))

Let 𝐻Sing
𝑘
+
(𝑀; 𝐺) denote the quotient group on the right-hand side of this equa-
tion. The last step of the proof is to show that 𝐻Sing
𝑘
+
𝑀; 𝐺) is isomorphic to
𝐻Sing
𝑘
(𝑀; 𝐺), which is given by the same formula but with Sing∗ and 𝛿 in place
of Sing∗,+ and 𝛿 + .
170 6. Sheaf Cohomology

Consider the following commutative diagram of group homomorphisms:

0 0 0

𝜃/
0 / Sing0 (𝑀; 𝐺)  
0
/ Sing0 (𝑀; 𝐺) Sing0,+ (𝑀; 𝐺) /0
  

𝛿 𝛿 𝛿+
𝜃/
0 / Sing1 (𝑀; 𝐺)  
0
/ Sing1 (𝑀; 𝐺) Sing1,+ (𝑀; 𝐺) /0
  
(6.23)
𝛿 𝛿 𝛿+
𝜃/
0 / Sing2 (𝑀; 𝐺)  
0
/ Sing2 (𝑀; 𝐺) Sing2,+ (𝑀; 𝐺) / 0,
  

⋮ ⋮ ⋮
  

where for each 𝑘, Sing𝑘0 (𝑀; 𝐺) is the subgroup of cochains 𝜑 ∈ Sing𝑘 (𝑀; 𝐺) that
are locally zero, meaning there is some open cover 𝒰 = {𝑈𝛼 }𝛼∈𝐴 such that 𝜑|𝑈𝛼 =
0 for every 𝛼; and 𝜃 = 𝜃Sing𝑘 is the global section map associated with the canonical
presheaf morphism 𝜃Sing𝑘 ∶ Sing𝑘 → Sing𝑘,+ given by Theorem 5.9. We begin by

Exactness at Sing𝑘0 (𝑀; 𝐺) and Sing𝑘 (𝑀; 𝐺) follows easily from the definitions.
showing that each horizontal row is exact.

To see that 𝜃 is surjective, suppose Φ ∈ Sing𝑘,+ (𝑀; 𝐺) is arbitrary. By Lemma

5.8, there is an open cover 𝒰 = {𝑈𝛼 }𝛼∈𝐴 of 𝑀 and for each 𝛼 a cochain 𝜑𝛼 ∈
Sing𝑘 (𝑈𝛼 ; 𝐺) such that Φ|𝑈𝛼 = 𝜑+ 𝛼 . By passing to a refinement, we may assume

on overlaps: since the presheaf Sing𝑘 does not satisfy the locality property, cochains
the cover is locally finite. It is important to note that these cochains might not agree

with the same germs at each point need not agree. Thus we need a more delicate

Let {𝜓𝛼 }𝛼∈𝐴 be a topological partition of unity subordinate to the cover 𝒰. For
argument.

each 𝑥 ∈ 𝑀, we wish to choose a neighborhood 𝑉𝑥 of 𝑥 satisfying the following

properties:
(i) The set 𝐴(𝑥) = {𝛼 ∈ 𝐴 ∶ 𝑉𝑥 ∩ supp 𝜓𝛼 ≠ ∅} is finite.
(ii) 𝑉𝑥 ⊆ 𝑈𝛼 for each 𝛼 ∈ 𝐴(𝑥).
(iii) The restrictions 𝜑𝛼 |𝑉𝑥 are all equal for 𝛼 ∈ 𝐴(𝑥).
(iv) 𝑉𝑥 ∩ supp 𝜓𝛽 = ∅ if 𝛽 ∉ 𝐴(𝑥).
To see that this is possible, note that the existence of a neighborhood 𝑉𝑥 satisfying
(i) is just a restatement of the local finiteness of the cover {supp 𝜓𝛼 }. Then we can
shrink 𝑉𝑥 successively to satisfy the other three conditions: (ii) because ⋂𝛼∈𝐴(𝑥) 𝑈𝛼
is an open set containing 𝑥; (iii) because the cochains 𝜑𝛼 all have the same germ at
𝑥 for 𝛼 ∈ 𝐴(𝑥); and (iv) because ⋃𝛽∉𝐴(𝑥) supp 𝜓𝛽 is closed in 𝑀 by local finiteness
Sheaf Cohomology and Singular Cohomology 171

[LeeTM, Lemma 4.75]. For each 𝑥 ∈ 𝑀, let 𝜑 ̃ 𝑥 ∈ Sing𝑘 (𝑉𝑥 ; 𝐺) be the cochain
that is the common value of 𝜑𝛼 |𝑉𝑥 for all 𝛼 ∈ 𝐴(𝑥).
Now suppose 𝑥 and 𝑦 are points of 𝑀 such that 𝑉𝑥 ∩ 𝑉𝑦 ≠ ∅, and let 𝑧 be a
point in 𝑉𝑥 ∩ 𝑉𝑦 . There is an index 𝛼 ∈ 𝐴 such that 𝑧 ∈ supp 𝜓𝛼 , and then property
(iv) ensures that 𝛼 lies in both 𝐴(𝑥) and 𝐴(𝑦). This implies that

(6.24) ̃ 𝑦 |𝑉𝑥 ∩𝑉𝑦 .

̃ 𝑥 |𝑉𝑥 ∩𝑉𝑦 = 𝜑𝛼 |𝑉𝑥 ∩𝑉𝑦 = 𝜑
𝜑

Thus we can define a global cochain 𝜑

̃ ∈ Sing𝑘 (𝑀; 𝐺) as follows: for each singular
𝑘-simplex 𝜎 in 𝑀, we set

𝜑
̃ 𝑥 (𝜎) if 𝜎(Δ𝑘 ) ⊆ 𝑉𝑥 for some 𝑥 ∈ 𝑀,
𝜑
̃ (𝜎) =
{0 if 𝜎(Δ𝑘 ) is not contained in any 𝑉𝑥 ,

and (6.24) ensures that this is well defined. It follows that 𝜃(̃
𝜑) = Φ, showing that
𝜃 ∶ Sing (𝑀; 𝐺) → Sing (𝑀; 𝐺) is surjective for each 𝑘.
𝑘 𝑘,+

The zigzag lemma applied to (6.23) yields a long exact sequence which reads
in part

𝜃∗
𝐻0𝑘 (𝑀; 𝐺) → 𝐻Sing
𝑘
(𝑀; 𝐺) −−→ 𝐻Sing
𝑘
+
(𝑀; 𝐺) → 𝐻0𝑘+1 (𝑀; 𝐺),

where 𝐻0𝑘 (𝑀; 𝐺) denotes the 𝑘th cohomology group of the leftmost column of
(6.23). Thus to complete the proof, it suffices to show that 𝐻0𝑘 (𝑀; 𝐺) = 0 for all 𝑘.
To that end, suppose 𝜑 ∈ Sing𝑘0 (𝑀; 𝐺) satisfies 𝛿𝜑 = 0. By definition of
Sing𝑘0 , there is an open cover 𝒰 = {𝑈𝛼 }𝛼∈𝐴 of 𝑀 such that 𝜑|𝑈𝛼 = 0 for each
𝛼. A singular simplex in 𝑀 is said to be 𝓤-small if its image is contained in
some 𝑈𝛼 , and a chain 𝑐 ∈ Sing𝑘 (𝑀) is 𝒰-small if it can be written as a for-
mal linear combination of 𝒰-small simplices. Let Sing𝒰 𝑘 (𝑀) ⊆ Sing𝑘 (𝑀) de-
note the subgroup consisting of 𝒰-small chains. The singular boundary operator
takes Sing𝒰 𝑘 (𝑀) to Sing𝑘−1 (𝑀), so we have a chain complex, which we denote
𝒰

by Sing𝒰 ∗ (𝑀). The inclusion maps 𝜄 ∶ Sing𝑘 (𝑀) → Sing𝑘 (𝑀) commute with
𝒰

the boundary operators and thus define a chain map from Sing𝒰 ∗ (𝑀) to Sing∗ (𝑀)
(where Sing∗ (𝑀) denotes the full singular chain complex), and a subdivision ar-

on all the homology groups. Let Sing𝑘,𝒰 (𝑀; 𝐺) = Hom ( Sing𝒰 𝑘 (𝑀), 𝐺), and let
gument [LeeTM, Prop. 13.19] shows that this chain map induces isomorphisms

𝜄∗ ∶ Sing𝑘 (𝑀; 𝐺) → Sing𝑘,𝒰 (𝑀; 𝐺) be the map dual to 𝜄; concretely, 𝜄∗ (𝜑) is just
the restriction of a cochain 𝜑 to 𝒰-small chains. Because 𝜄 induces isomorphisms
on all homology groups, it follows from Proposition 6.17 that 𝜄∗ induces isomor-
phisms on cohomology.
172 6. Sheaf Cohomology

Let Sing𝑘,𝒰
0
(𝑀; 𝐺) ⊆ Sing𝑘 (𝑀; 𝐺) denote the kernel of 𝜄∗ ; it is the group of
cochains that assign the value zero to every 𝒰-small simplex. We have a commu-
tative diagram
0 0 0

𝜄∗ /
0 / Sing0,𝒰 (𝑀; 𝐺)   / Sing0 (𝑀; 𝐺) Sing0,𝒰 (𝑀; 𝐺) /0
  
0

𝛿 𝛿 𝛿
𝜄∗ /
0 / Sing1,𝒰 (𝑀; 𝐺) / Sing1 (𝑀; 𝐺) Sing1,𝒰 (𝑀; 𝐺) / 0,
  
0
𝛿 𝛿 𝛿


⋮ ⋮ ⋮
  

in which the horizontal rows are exact. Thus the zigzag lemma yields a long exact
cohomology sequence, which reads in part
≅
𝐻Sing
𝑘−1
(𝑀; 𝐺) ⟶𝐻Sing,𝒰
𝑘−1
(𝑀; 𝐺) → 𝐻Sing,𝒰,0
𝑘
(𝑀; 𝐺)
≅
→ 𝐻Sing
𝑘
(𝑀; 𝐺) ⟶ 𝐻Sing,𝒰
𝑘
(𝑀; 𝐺),

where the notations 𝐻Sing,𝒰,0

𝑘
and 𝐻Sing,𝒰
𝑘

that the first and last maps are isomorphisms implies that 𝐻Sing,𝒰,0
have the obvious meanings. The fact
𝑘
(𝑀; 𝐺) = 0.
Therefore, there is a cochain 𝛽 ∈ Sing𝑘−1,𝒰
0
(𝑀; 𝐺) ⊆ Sing𝑘−1
0
(𝑀; 𝐺) such that
𝛿𝛽 = 𝜑. This completes the proof that 𝐻 (𝑀; 𝐺) ≅ 𝐻Sing (𝑀; 𝐺).
𝑘 𝑘

To prove the naturality statement, we note that the isomorphism 𝐻 𝑘 (𝑀; 𝐺) ≅

𝐻Sing
𝑘
(𝑀; 𝐺) is the composition of two isomorphisms:

𝐻 𝑘 (𝑀; 𝐺) ≅ 𝐻Sing
𝑘
+
(𝑀; 𝐺) ≅ 𝐻Sing
𝑘
(𝑀; 𝐺).
The first isomorphism was obtained by applying the de Rham–Weil theorem to
the sequence (6.22), so it is natural with respect to group homomorphisms by the

induced by the map 𝜃 ∶ Sing𝑘 (𝑀; 𝐺) → Sing𝑘,+ (𝑀; 𝐺), which just maps a cochain
naturality statement of the de Rham–Weil theorem. The second isomorphism is

𝜑 to the section 𝜑+ of the sheafification defined by (5.6); and it is straightforward

to check that this map is also natural with respect to group homomorphisms. □

We have proved the preceding theorem for locally contractible paracompact

Hausdorff spaces, which includes all topological manifolds with or without bound-
ary and is more than adequate for our purposes. But the conclusion is true for
somewhat more general spaces: an unpublished 2016 paper by Yehonatan Sella

𝑀 that is semilocally contractible, meaning that each open subset 𝑈 ⊆ 𝑀 has an

[Sel16] (see also [Pet22]) shows that it is true for constant sheaves over any space
Applications of Sheaf Cohomology 173

open cover by sets whose inclusions into 𝑈 are homotopic to constant maps, with-

however: for example, if 𝑀 is a paracompact Hausdorff space that is connected but

out any paracompactness requirement. Some topological restriction is necessary,

not path-connected, then 𝐻 0 (𝑀; ℤ) ≅ ℤ, while 𝐻Sing

0
(𝑀; ℤ) is a direct product of
copies of ℤ, one for each path component.

Applications of Sheaf Cohomology

In this section, we describe several applications of sheaf cohomology to complex
manifolds.

The Dolbeault Theorem

Our first application is central to complex manifold theory. This theorem was
first proved in the 1950s by Pierre Dolbeault [Dol53, Dol56].
Theorem 6.19 (The Dolbeault Theorem). Suppose 𝑀 is a complex manifold and
Ω𝑝 is its sheaf of holomorphic 𝑝-forms. For each 𝑞 ≥ 0,
(6.25) 𝐻 𝑝,𝑞 (𝑀) ≅ 𝐻 𝑞 (𝑀; Ω𝑝 ).
More generally, if 𝐸 → 𝑀 is a holomorphic vector bundle, then
(6.26) 𝐻 𝑝,𝑞 (𝑀; 𝐸) ≅ 𝐻 𝑞 (𝑀; Ω𝑝 (𝐸)).

Proof. The isomorphism (6.25) is just the special case of (6.26) in which 𝐸 is the
trivial line bundle 𝑀 × ℂ → 𝑀, so we just need to prove (6.26). Example 5.23(c)
showed that the sheaves of smooth 𝐸-valued forms ℰ 𝑝,𝑞 (𝐸) give a resolution of
Ω𝑝 (𝐸). Because each sheaf ℰ 𝑝,𝑞 (𝐸) is a sheaf of ℰ -modules, it is fine and therefore
acyclic. The theorem follows from the de Rham–Weil theorem. □

The de Rham Theorem

As we mentioned above, the de Rham–Weil theorem can be used to give a quick
sheaf-theoretic proof of the following version of the de Rham theorem.
Theorem 6.20 (De Rham Theorem, Sheaf-Theoretic Version). Suppose 𝑀 is a
smooth manifold. For each 𝑘 ≥ 0,
(6.27) 𝐻dR
𝑘
(𝑀; ℂ) ≅ 𝐻 𝑘 (𝑀; ℂ).
The analogous statement is true for real de Rham cohomology.

Proof. For each nonnegative integer 𝑘, let ℰ 𝑘 be the sheaf of smooth complex-
valued 𝑘-forms on 𝑀. Consider the following sheaf sequence:
𝑑 𝑑 𝑑
(6.28) 0 → ℂ ↪ ℰ0 −
→ ℰ1 −
→ ℰ2 −
→ ⋯.
174 6. Sheaf Cohomology

ℰ 𝑘 are fine, it is an acyclic resolution of the constant sheaf ℂ. Thus the cohomology
We observed in Example 5.23(a) that this sequence is exact, and since the sheaves

groups) are isomorphic to the sheaf cohomology groups of ℂ by the de Rham–Weil

groups of its global section sequence (which are the complex de Rham cohomology

theorem. The same argument applies with real coefficients. □

The naturality statement of the de Rham–Weil theorem leads to the following

relationship between the de Rham and Dolbeault cohomology groups on a complex
manifold.
Proposition 6.21. Let 𝑀 be a complex manifold. For each integer 𝑞 ≥ 0, the pro-
jection 𝜋 0,𝑞 ∶ ℰ 𝑞 (𝑀) → ℰ 0,𝑞 (𝑀) descends to a linear map 𝜋∗ ∶ 𝐻dR (𝑀; ℂ) →
0,𝑞 𝑞

𝐻 0,𝑞 (𝑀), such that the following diagram commutes:

ℛ / 𝑞
𝐻dR (𝑀; ℂ) 𝐻 (𝑀; ℂ)
𝑞

(6.29) 𝜋∗
0,𝑞 𝑖∗

𝐻 0,𝑞 (𝑀) / 𝐻 𝑞 (𝑀; 𝒪),

𝒟
 

where ℛ and 𝒟 are the isomorphisms given by Theorems 6.20 and 6.19, respec-
tively, and 𝑖∗ is induced by the sheaf inclusion 𝑖 ∶ ℂ ↪ 𝒪.

Proof. Consider the following diagram of sheaf morphisms:

𝑑 𝑑
0 /ℂ / ℰ0 / ℰ1 / ℰ2 /⋯

𝑖 = 𝜋 0,1 𝜋 0,2
𝜕 / 0,1 𝜕 / 0,2
0 /𝒪 / ℰ 0,0 ℰ ℰ / ⋯.
   

For 𝜂 ∈ ℰ 𝑞 (𝑈 ) on some open subset 𝑈 ⊆ 𝑀, we have a decomposition

𝜂 = 𝜂 𝑞,0 + 𝜂 𝑞−1,1 + ⋯ + 𝜂 1,𝑞−1 + 𝜂 0,𝑞 .
The only part of 𝜂 that can contribute to the (0, 𝑞 + 1)-part of 𝑑𝜂 is 𝜂 0,𝑞 , so
𝜋 0,𝑞+1 𝑑𝜂 = 𝜕𝜋 0,𝑞 𝜂. Thus the above sheaf diagram commutes, and it follows that
𝜋 0,𝑞 descends to a well-defined linear map from 𝐻dR (𝑀; ℂ) to 𝐻 0,𝑞 (𝑀). Commu-
𝑞

tativity of (6.29) is an immediate consequence of the naturality statement of the de

Rham–Weil theorem. □

It is important to note that the above proof does not apply to 𝜋 𝑝,𝑞 for 𝑝 ≠ 0, be-
cause the (𝑝, 𝑞 +1)-part of 𝑑𝜂 can involve contributions from both 𝜂 𝑝,𝑞 and 𝜂 𝑝−1,𝑞+1 .

there are well-defined projections 𝜋∗ ∶ 𝐻 𝑝+𝑞 (𝑀; ℂ) → 𝐻 𝑝,𝑞 (𝑀), but the proof is
𝑝,𝑞
Later we will see that for a restricted class of manifolds (compact Kähler manifolds),

considerably more complicated (see Thm. 9.44).

Applications of Sheaf Cohomology 175

By combining the isomorphism 𝐻dR𝑘

(𝑀; ℝ) ≅ 𝐻 𝑘 (𝑀; ℝ) of Theorem 6.20
with the isomorphism 𝐻 𝑘 (𝑀; ℝ) ≅ 𝐻Sing
𝑘
(𝑀; ℝ) of Theorem 6.18, we obtain the
classical de Rham isomorphism 𝐻dR 𝑘
(𝑀; ℝ) ≅ 𝐻Sing
𝑘
(𝑀; ℝ) and its complex ana-
logue. But we need a more precise version of this, which says explicitly that the
isomorphism is given by integration of differential forms. For that purpose, we

Here is the setup. Suppose 𝑀 is a smooth manifold and 𝑘 is a nonnegative

introduce a modified version of singular cohomology.

integer. A singular 𝑘-simplex 𝜎 ∶ Δ𝑘 → 𝑀 is called a smooth singular simplex if

borhood of Δ𝑘 in ℝ𝑘 . A chain 𝑐 ∈ Sing𝑘 (𝑀) is called a smooth (singular) chain if

it is smooth in the sense that it can be extended to a smooth map defined on a neigh-

Sing∞ 𝑘 (𝑀) ⊆ Sing𝑘 (𝑀) denote the subgroup consisting of smooth singular chains
it can be written as a formal linear combination of smooth singular simplices. Let

in 𝑀. Since the singular boundary operator maps smooth chains to smooth chains,
the sequence
𝜕 𝜕
⋯ → Sing∞
𝑘+1 (𝑀) → Sing𝑘 (𝑀) → Sing𝑘−1 (𝑀) → ⋯
∞ ∞

groups and denoted by 𝐻𝑘∞ (𝑀).

is a chain complex, whose homology groups are called smooth singular homology

The fundamental fact about smooth singular homology is the following theo-
rem, whose proof can be found in [LeeSM, Thm. 18.7].

manifold 𝑀 and nonnegative integer 𝑘, the map 𝑖∗ ∶ 𝐻𝑘∞ (𝑀) → 𝐻𝑘 (𝑀) induced
Theorem 6.22 (Smooth Singular vs. Singular Homology). For every smooth

𝑘 (𝑀) ↪ Sing𝑘 (𝑀) is an isomorphism.

by inclusion 𝑖 ∶ Sing∞
Now let 𝐺 be an abelian group and let
Sing𝑘,∞ (𝑀; 𝐺) = Hom ( Sing∞
𝑘 (𝑀), 𝐺).
The singular coboundary maps 𝛿 ∶ Sing𝑘,∞ (𝑀; 𝐺) → Sing𝑘+1,∞ (𝑀; 𝐺) are de-
fined just as in the case of continuous cochains, and they satisfy 𝛿 ∘ 𝛿 = 0. Thus we
have a cochain complex
𝛿 𝛿
⋯ → Sing𝑘−1,∞ (𝑀; 𝐺) → Sing𝑘,∞ (𝑀; 𝐺) → Sing𝑘+1,∞ (𝑀; 𝐺) → ⋯ .
Its cohomology groups, which we denote by 𝐻Sing,∞
𝑘
(𝑀; 𝐺), are called the smooth
singular cohomology groups of 𝑴 with coefficients in 𝑮,
The dual of the inclusion map 𝑖 ∶ Sing∞ 𝑘 (𝑀) ↪ Sing𝑘 (𝑀) is a homomor-
phism 𝑅 ∶ Sing𝑘 (𝑀; 𝐺) → Sing𝑘,∞ (𝑀; 𝐺), which is just restriction: if 𝜑 is a
singular 𝑘-cochain on 𝑀, then 𝑅(𝜑) is just the restriction of 𝜑 to smooth singular

map 𝑅∗ ∶ 𝐻Sing
𝑘
(𝑀; 𝐺) → 𝐻Sing,∞
𝑘
(𝑀; 𝐺). Because inclusion of smooth chains
chains. It commutes with the coboundary operators and thus descends to a linear

induces isomorphisms in homology, the next corollary follows immediately from

Proposition 6.17.
176 6. Sheaf Cohomology

Corollary 6.23 (Smooth Singular vs. Singular Cohomology). Suppose 𝑀 is a

smooth manifold and 𝐺 is an abelian group or a real or complex vector space. The
map 𝑅∗ ∶ 𝐻Sing
𝑘
(𝑀; 𝐺) → 𝐻Sing,∞
𝑘
(𝑀; 𝐺) is an isomorphism for each nonnegative
integer 𝑘.
Let ℰ 𝑘 (𝑀) denote the space of smooth complex-valued 𝑘-forms on 𝑀. If
□

𝜂 ∈ ℰ 𝑘 (𝑀) and 𝑐 = ∑𝑗 𝑎𝑗 𝜎𝑗 is a smooth singular 𝑘-chain, we define the integral

of 𝜼 over 𝒄 by

𝜂= 𝑎𝑗 𝜎𝑗∗ 𝜂.
∫𝑐
∑ ∫Δ𝑘
𝑗

Define a linear map 𝐼 ∶ ℰ 𝑘 (𝑀) → Sing𝑘,∞ (𝑀; ℂ) by

𝐼(𝜂)(𝑐) = 𝜂
∫𝑐

for every smooth 𝑘-form 𝜂 and smooth chain 𝑐. A version of Stokes’s theorem for
smooth singular chains [LeeSM, Thm. 18.12] shows that

𝑑𝜂 = 𝜂.
∫ ∫
(6.30)
𝑐 𝜕𝑐

This implies

𝛿(𝐼(𝜂))(𝑐) = 𝐼(𝜂)(𝜕𝑐) = 𝜂= 𝑑𝜂 = 𝐼(𝑑𝜂)(𝑐),

∫𝜕𝑐 ∫𝑐

which is to say that

(6.31) 𝛿 ∘ 𝐼 = 𝐼 ∘ 𝑑.

Thus 𝐼 descends to a linear map 𝐼∗ ∶ 𝐻dR

𝑘
(𝑀; ℂ) → 𝐻Sing,∞
𝑘
(𝑀; ℂ).
Here is the theorem that describes the detailed relationship between de Rham
and singular cohomology. (For a different proof, see [LeeSM, Thm. 18.14].)

manifold 𝑀 and every integer 𝑘 ≥ 0, the map 𝐼∗ ∶ 𝐻dR

𝑘
(𝑀; ℂ) → 𝐻Sing,∞
𝑘
(𝑀; ℂ)
Theorem 6.24 (De Rham Theorem, Integration Version). For every smooth

is an isomorphism. The analogous statement holds for real de Rham cohomology.

just treat the complex case. Let 𝑀 be a smooth manifold. For each nonnegative
Proof. The proofs for real and complex coefficients are essentially identical, so we

integer 𝑘, the assignment 𝑈 ↦ Sing𝑘,∞ (𝑈 ; ℂ) is a presheaf on 𝑀, and there is a

presheaf morphism 𝐼 ∶ ℰ 𝑘 → Sing𝑘,∞ , with 𝐼 ∶ ℰ 𝑘 (𝑈 ) → Sing𝑘,∞ (𝑈 ; ℂ) defined
by integration over smooth chains in 𝑈 as above. Let Sing𝑘,∞+ be the sheafification
of the presheaf Sing𝑘,∞ , and let 𝐼 + ∶ ℰ 𝑘 → Sing𝑘,∞+ be the sheafification of 𝐼
(where we identify the sheaf ℰ 𝑘 with its sheafification). Consider the following
Applications of Sheaf Cohomology 177

diagram of sheaf morphisms:

𝑑 𝑑 𝑑
0 /ℂ / ℰ0 / ℰ1 / ℰ2 /⋯

(6.32) = 𝐼+ 𝐼+ 𝐼+
0 /ℂ / Sing0,∞+ / Sing1,∞+ / Sing2,∞+ / ⋯.
𝛿 𝛿 𝛿+
  
+ +


is an acyclic resolution of ℂ. The proof of Theorem 6.18 showed that the analogue
It commutes by virtue of (6.31) and the functoriality of sheafification. The top row

of the second row, with Sing𝑘,+ in place of Sing𝑘,∞+ , is also an acyclic resolution;
the same proof applies in this case, once we observe that the smooth singular coho-
mology groups of a contractible open set are trivial by virtue of Theorem 6.22 and
the universal coefficient theorem.

find that the following diagram commutes for each 𝑘:

Applying the naturality statement of the de Rham–Weil theorem to (6.32), we

≅
𝐻 𝑘 (𝑀; ℂ) / 𝐻 𝑘 (𝑀; ℂ)

= (𝐼 + )∗
dR

𝐻 𝑘 (𝑀; ℂ) / 𝐻𝑘 (𝑀; ℂ).

≅
 
Sing,∞+

Thus (𝐼 + )∗ is an isomorphism. On the other hand, Theorem 5.9 shows there is a

commutative sheaf diagram
𝐼
ℰ𝑘 / Sing𝑘,∞

= 𝜃
ℰ𝑘 / Sing𝑘,∞+
𝐼+
 

(still identifying ℰ 𝑘 with its sheafification), and passing to induced cohomology

maps, we find that the isomorphism (𝐼 + )∗ is equal to the composition
𝐼∗ 𝜃∗
(6.33) 𝐻dR
𝑘
(𝑀; ℂ) −−→ 𝐻Sing,∞
𝑘
(𝑀; ℂ) −−→ 𝐻Sing,∞+
𝑘
(𝑀; ℂ).
Thus to show that 𝐼∗ is an isomorphism, we need to show that the map 𝜃∗ above

𝜃∗ ∶ 𝐻Sing
𝑘
(𝑀; ℂ) → 𝐻Sing,+
𝑘
(𝑀; ℂ) is an isomorphism. The same proof applies
is an isomorphism. The proof of Theorem 6.18 showed that the analogous map

here; the only additional observation that needs to be made is that when the subdivi-
sion operators of [LeeTM, Prop. 13.19] are applied to smooth chains, they produce
smooth chains, because the new singular simplices are defined as compositions
of the original smooth simplices with affine maps between subsets of Euclidean
spaces. □
178 6. Sheaf Cohomology

The next corollary is an immediate consequence of Corollary 6.23 and Theorem

6.24.
Corollary 6.25. For a smooth manifold 𝑀 and a nonnegative integer 𝑘, define
ℐ ∶ 𝐻dR𝑘
(𝑀; ℂ) → 𝐻Sing
𝑘
(𝑀; ℂ) by ℐ = 𝑅−1 ∗ ∘𝐼∗ , where 𝐼∗ is the map of Theorem
6.24 and 𝑅∗ is the map of Corollary 6.23. Then ℐ is an isomorphism.
Corollary 6.26. Let 𝑀 be a smooth manifold. For any closed 𝑘-form 𝜂 and
□

singular 𝑘-cycle 𝑐, the Kronecker pairing between ℐ [𝜂] ∈ 𝐻Sing

𝑘
(𝑀; ℂ) and
[𝑐] ∈ 𝐻𝑘 (𝑀) is given by
⟨ℐ [𝜂], [𝑐]⟩ = 𝜂,
∫𝑐̃
where 𝑐 ̃ is any smooth cycle cohomologous to 𝑐.

Proof. Let 𝜂 and 𝑐 be given. By definition, ℐ [𝜂] = 𝑅−1 ∗ ∘ 𝐼∗ [𝜂]. Thus there
is a singular cocycle 𝜑 ∈ Sing𝑘 (𝑀; ℂ) such that 𝑅(𝜑) ∈ Sing𝑘,∞ (𝑀; ℂ) is co-
homologous to 𝐼(𝜂), and ℐ [𝜂] = [𝜑]. By definition of the Kronecker pairing,
⟨ℐ [𝜂], [𝑐]⟩ = 𝜑(𝑐). Theorem 6.22 shows that there is a smooth cycle 𝑐 ̃ homol-
ogous to 𝑐, and since the cocycle 𝜑 gives the same value on homologous cycles,
the Kronecker pairing is also equal to 𝜑(𝑐).
̃ Since 𝑐 ̃ is smooth, this is the same as
𝑅(𝜑)(𝑐)̃ = 𝐼(𝜂)(𝑐)̃ = ∫𝑐 ̃ 𝜂. □

resent integral cohomology classes in the following sense: If 𝑀 is a smooth

In Chapters 9 and 10, it will be important to know which closed forms rep-

manifold, we say that a cohomology class in 𝐻dR 𝑘

(𝑀; ℂ) (or either of the iso-
morphic groups 𝐻 (𝑀; ℂ) or 𝐻Sing (𝑀; ℂ)) is integral if it lies in the image of
𝑘 𝑘

the coefficient homomorphism 𝑖∗ ∶ 𝐻Sing

𝑘
(𝑀; ℤ) → 𝐻Sing
𝑘
(𝑀; ℂ) (or equivalently
𝐻 𝑘 (𝑀; ℤ) → 𝐻 𝑘 (𝑀; ℂ)) induced by inclusion 𝑖 ∶ ℤ ↪ ℂ.
The next lemma gives a simple criterion for detecting when a de Rham coho-
mology class is integral.
Lemma 6.27. Let 𝑀 be a smooth manifold and let 𝜂 be a closed 𝑘-form on 𝑀.
Then 𝜂 represents an integral cohomology class if and only if ∫𝑐 𝜂 ∈ ℤ for every
smooth 𝑘-cycle 𝑐 in 𝑀. If 𝐻𝑘 (𝑀) is finitely generated, then the set of integral
classes is a free abelian subgroup of 𝐻dR
𝑘
(𝑀; ℂ), a basis for which is also a basis
for 𝐻dR (𝑀; ℂ) over ℂ and a basis for 𝐻dR (𝑀; ℝ) over ℝ.
𝑘 𝑘

Proof. Consider the following commutative diagram of group homomorphisms:

𝜅ℤ
𝐻Sing
𝑘
(𝑀; ℤ) / Hom(𝐻 (𝑀), ℤ)
𝑘

𝑖∗ 𝑖#

𝐻dR
𝑘
(𝑀; ℂ) / 𝐻 𝑘 (𝑀; ℂ)
𝜅ℂ Hom(𝐻𝑘 (𝑀), ℂ),
ℐ
 
/
Sing
Applications of Sheaf Cohomology 179

where 𝜅ℤ and 𝜅ℂ are the respective Kronecker homomorphisms and 𝑖# is post-

composition with 𝑖 ∶ ℤ ↪ ℂ. Corollary 6.26 shows that the composition 𝜅ℂ ∘ ℐ [𝜂]
is the homomorphism that sends [𝑐] ∈ 𝐻𝑘 (𝑀) to ∫𝑐 ̃ 𝜂 for a smooth cycle 𝑐 ̃ homol-
ogous to 𝑐.
On the one hand, if [𝜂] is integral, then there is a class 𝛾 ∈ 𝐻Sing
𝑘
(𝑀; ℤ) such
that 𝑖∗ 𝛾 = ℐ [𝜂], which implies
𝜅ℂ ∘ ℐ [𝜂] = 𝜅ℂ ∘ 𝑖∗ 𝛾 = 𝑖# ∘ 𝜅ℤ 𝛾,
and thus the homomorphism 𝜅ℂ ∘ ℐ [𝜂] takes only integer values.
Conversely, if 𝜅ℂ ∘ ℐ [𝜂] takes only integer values, then 𝜅ℂ ∘ ℐ [𝜂] = 𝑖# 𝜑 for
𝜑 ∈ Hom(𝐻𝑘 (𝑀); ℤ) (given by the same formula). Since 𝜅ℤ is surjective by
the universal coefficient theorem (Prop. 6.16), it follows that there is some 𝛾 ∈
𝐻Sing
𝑘
(𝑀; ℤ) such that 𝜅ℤ 𝛾 = 𝜑, which implies
𝜅ℂ ∘ ℐ [𝜂] = 𝑖# ∘ 𝜅ℤ 𝛾 = 𝜅ℂ ∘ 𝑖∗ 𝛾.
Since 𝜅ℂ is an isomorphism, this implies ℐ [𝜂] = 𝑖∗ 𝛾, so [𝜂] is integral.
To prove the last statement, assume 𝐻𝑘 (𝑀) is finitely generated, and let 𝑇 ⊆
𝐻𝑘 (𝑀) be its torsion subgroup (the subgroup of elements 𝑥 that satisfy 𝑛𝑥 = 0 for
some positive integer 𝑛). Then 𝐻𝑘 (𝑀)/𝑇 ≅ ℤ𝑚 for some nonnegative integer 𝑚.
Let (𝛾1 , … , 𝛾𝑚 ) be elements of 𝐻𝑘 (𝑀) whose images in the quotient group form a
basis for 𝐻𝑘 (𝑀)/𝑇 . Since every homomorphism to ℤ annihilates torsion elements,
Hom(𝐻𝑘 (𝑀), ℤ) ≅ Hom(𝐻𝑘 (𝑀)/𝑇 , ℤ), so it is a free abelian group with basis
(𝜑1 , … , 𝜑𝑚 ), where 𝜑𝑖 (𝛾𝑗 ) = 𝛿𝑗𝑖 . These same elements are also a basis for the
vector space Hom(𝐻𝑘 (𝑀), ℂ) ≅ Hom(𝐻𝑘 (𝑀)/𝑇 , ℂ) over ℂ, and therefore their
images in the isomorphic space 𝐻dR 𝑘
(𝑀; ℂ) are a basis over ℤ for the subgroup of
integral classes and also a basis over ℂ for 𝐻dR 𝑘
(𝑀; ℂ) itself. The argument for
𝐻dR𝑘
(𝑀; ℝ) is the same. □

Chern Classes of Line Bundles

For our next application, suppose 𝑀 is a complex manifold and 𝒪 ∗ is its sheaf
of nonvanishing holomorphic functions. Let 𝒰 = {𝑈𝛼 }𝛼∈𝐴 be an open cover of
𝑀. The computations in Example 6.3 showed that a Čech 1-cocycle on 𝒰 with
coefficients in 𝒪 ∗ consists of a collection of nonvanishing holomorphic functions
𝜏𝛼𝛽 ∶ 𝑈𝛼 ∩ 𝑈𝛽 → ℂ that satisfy
𝜏𝛼𝛾 = 𝜏𝛼𝛽 𝜏𝛽𝛾 on 𝑈𝛼 ∩ 𝑈𝛽 ∩ 𝑈𝛾 .
This is exactly the data for constructing a holomorphic line bundle, so we get a map
ℒ ∶ 𝑍 1 (𝒰; 𝒪 ∗ ) → Pic(𝑀)
where 𝑍 1 (𝒰; 𝒪 ∗ ) is the group of Čech cycles on 𝒰 with coefficients in 𝒪 ∗ , and
Pic(𝑀) is the Picard group of 𝑀 (the group of isomorphism classes of holomorphic
line bundles under the tensor product).
180 6. Sheaf Cohomology

Proposition 6.28. Let 𝑀 be a complex manifold. The map ℒ defined above de-
scends to a group isomorphism
ℒ∗ ∶ 𝐻 1 (𝑀; 𝒪 ∗ ) ≅ Pic(𝑀).
Similarly, if 𝑀 is a smooth manifold, 𝐻 1 (𝑀; ℰ ∗ ) is isomorphic to the group of
isomorphism classes of smooth complex line bundles on 𝑀; and if 𝑀 is any topo-
logical space, 𝐻 1 (𝑀; 𝒞 ∗ ) is isomorphic to the group of isomorphism classes of
topological complex line bundles.

Proof. We first note that for a fixed open cover 𝒰 of 𝑀, the map ℒ is a group
homomorphism from 𝑍 1 (𝒰; 𝒪 ∗ ) to Pic(𝑀), because the transition functions for a
tensor product 𝐿 ⊗ 𝐿′ are the products of the ones for 𝐿 and the ones for 𝐿′ .
Next we show that for a cocycle 𝜏 ∈ 𝑍 1 (𝒰; 𝒪 ∗ ), the isomorphism class of the
bundle ℒ (𝜏) depends only on the cohomology class of 𝜏. If 𝜏 and 𝜏 ′ are coho-
mologous, the computations in Example 6.3 show that there is a 0-cocycle 𝜓 such

𝜏𝛼𝛽 = 𝜏𝛼𝛽 𝜓𝛽 𝜓𝛼−1 on 𝑈𝛼 ∩ 𝑈𝛽 ,

that
′

so Proposition 3.7 shows that the bundles ℒ (𝜏) and ℒ (𝜏 ′ ) are isomorphic. Thus
ℒ descends to a homomorphism from 𝐻 1 (𝒰; 𝒪 ∗ ) to Pic(𝑀).
If 𝒱 = {𝑉𝛽 ∶ 𝛽 ∈ 𝐵} is a refinement of 𝒰 and 𝜌 is a refining map, then we
obtain a trivialization of the bundle ℒ (𝜏) over each subset 𝑉𝛽 by restricting the one
over 𝑈𝜌(𝛽) , and the transition maps between these trivializations are the restrictions
of those over 𝒰, which is to say
𝜏𝛼𝛽 = 𝜏𝜌(𝛼)𝜌(𝛽) |𝑉 .
𝛼 ∩𝑉𝛽

This implies that 𝒱 is a trivializing cover for both ℒ (𝜏) and ℒ (𝜌# 𝜏) with the same
transition functions, so these two bundles are isomorphic. Therefore, ℒ passes to
the direct limit to define a homomorphism ℒ∗ ∶ 𝐻 1 (𝑀; 𝒪 ∗ ) → Pic(𝑀).
If 𝐿 → 𝑀 is any holomorphic line bundle, we can choose a trivializing cover
𝒰, and the argument above shows that the transition functions determine a 1-
cocycle and therefore an element of 𝐻 1 (𝑀; 𝒪 ∗ ) whose image under ℒ∗ is the iso-
morphism class of 𝐿; thus ℒ∗ is surjective.
On the other hand, to show that ℒ∗ is injective, it suffices to show that its kernel
consists only of the zero cohomology class. Suppose ℒ∗ (𝛾) is the equivalence class
of the trivial bundle for some 𝛾 ∈ 𝐻 1 (𝑀; 𝒪 ∗ ). If 𝜏 is a cocycle representing 𝛾 over
some open cover 𝒰, then Corollary 3.9 shows there is a 0-cochain 𝜓 ∈ 𝐶 0 (𝒰; 𝒪 ∗ )
such that 𝜏𝛼𝛽 = 𝜓𝛽 𝜓𝛼−1 on 𝑈𝛼 ∩ 𝑈𝛽 , which means 𝜏 = 𝛿𝜓. Thus 𝜏 represents the
trivial cohomology class, which is to say 𝛾 = 0.
The analogous results for smooth and topological line bundles are proved in
exactly the same way, with only minor changes in notation. □
Applications of Sheaf Cohomology 181

Suppose 𝑀 is a smooth manifold. Consider the smooth exponential sheaf se-

quence on 𝑀 (Example 5.23(e)):
𝜀
(6.34) 0 → ℤ ↪ ℰ ⟶ ℰ ∗ → 0.

homomorphism 𝛿∗ ∶ 𝐻 1 (𝑀; ℰ ∗ ) → 𝐻 2 (𝑀; ℤ). By Proposition 6.28, 𝐻 1 (𝑀; ℰ ∗ )

The connecting homomorphism of the associated long exact sequence is a group

Given such a bundle 𝐿, we define the (sheaf-theoretic) Chern class of 𝑳 to be

is isomorphic to the group of isomorphism classes of smooth complex line bundles.

𝑐(𝐿) = −𝛿∗ ([𝐿]) ∈ 𝐻 2 (𝑀; ℤ), or its image in 𝐻Sing

2
(𝑀; ℤ) under the isomorphism

explained later. Some authors define the sheaf-theoretic Chern class as 𝑐(𝐿) =
given by Theorem 6.18. (The negative sign is a normalization constant that will be

𝛿 ∗ ([𝐿]), without the negative sign; see the remark following the proof of Theorem
7.14 below.)
Theorem 6.29 (Classification of Smooth Line Bundles). Let 𝑀 be a smooth man-
ifold. Smooth complex line bundles over 𝑀 are classified up to isomorphism by
their Chern classes: for every cohomology class 𝛾 ∈ 𝐻 2 (𝑀; ℤ), there is a smooth
complex line bundle 𝐿 with 𝑐(𝐿) = 𝛾, and two smooth complex line bundles are
smoothly isomorphic if and only if their Chern classes are equal.

Proof. The long exact sequence associated with (6.34) contains the following seg-

𝛿∗
ment:
𝐻 1 (𝑀; ℰ ) → 𝐻 1 (𝑀; ℰ ∗ ) ⟶ 𝐻 2 (𝑀; ℤ) → 𝐻 2 (𝑀; ℰ ).
Because ℰ is a fine sheaf, the leftmost and rightmost groups are zero, so −𝑐 =
𝛿∗ ∶ 𝐻 1 (𝑀; ℰ ∗ ) → 𝐻 2 (𝑀; ℤ) is an isomorphism. □

happens. Let 𝑀 be a complex manifold, and consider the holomorphic version of

When we turn our attention to holomorphic line bundles, something different

the exponential sequence:

𝜀
(6.35) 0 → ℤ ↪ 𝒪 ⟶ 𝒪 ∗ → 0.
Now the long exact sequence reads
𝛿∗
(6.36) 𝐻 1 (𝑀; 𝒪) → 𝐻 1 (𝑀; 𝒪 ∗ ) ⟶ 𝐻 2 (𝑀; ℤ) → 𝐻 2 (𝑀; 𝒪).
By Proposition 6.28, 𝐻 1 (𝑀; 𝒪 ∗ ) is isomorphic to the Picard group of 𝑀. Again,
we define the Chern class of a holomorphic line bundle 𝐿 by 𝑐(𝐿) = −𝛿∗ ([𝐿]). By
naturality of the long exact sequence applied to the following commutative diagram
𝜀 / ∗
0 / ℤ  / 𝒪 𝒪 _ /0

=
_

0 / ℤ  /ℰ
𝜀
/ ℰ∗ / 0,
  
182 6. Sheaf Cohomology

the following diagram commutes:

𝛿∗
𝐻 1 (𝑀; 𝒪 ∗ ) / 𝐻 2 (𝑀; ℤ)

𝐻 1 (𝑀; ℰ ∗ ) / 𝐻 2 (𝑀; ℤ),

𝛿∗
 

and therefore 𝑐(𝐿) depends only on the smooth isomorphism class of 𝐿.

phic if and only if their Chern classes are equal. However, since 𝒪 is not fine,
By the previous theorem, two holomorphic line bundles are smoothly isomor-

map 𝑐 ∶ Pic(𝑀) → 𝐻 2 (𝑀; ℤ) might not be injective and/or surjective. Thus if

it need not be the case that the groups on the ends of (6.36) are zero, so the

𝐻 2 (𝑀; 𝒪) ≠ 0, there might be smooth line bundles on 𝑀 that admit no holomor-

phic structures; and if 𝐻 1 (𝑀; 𝒪) ≠ 0, there might be smooth line bundles that
admit multiple nonisomorphic holomorphic structures.

𝑴 and denoted by Pic0 (𝑀). It is the group of isomorphism classes of holomorphic

The kernel of the Chern class homomorphism is called the Picard variety of

line bundles with zero Chern class, or equivalently holomorphic structures on the
trivial smooth line bundle. In Chapter 9, we will see that for a large class of compact

surfaces, Pic0 (𝑀) has the structure of a complex torus (see Thm. 9.66).
complex manifolds, including all projective manifolds and all compact Riemann

Line Bundles on Riemann Surfaces

𝑀 is a connected compact Riemann surface. For any smooth complex line bundle
For line bundles on compact Riemann surfaces, we can say more. Suppose

𝐿 → 𝑀, the degree of 𝑳 is defined to be the integer

deg(𝐿) = ⟨𝑐(𝐿), [𝑀]⟩,

where 𝑐(𝐿) ∈ 𝐻Sing

2
(𝑀; ℤ) is the Chern class of 𝐿, [𝑀] ∈ 𝐻2 (𝑀) is the funda-
mental class of 𝑀, and ⟨⋅, ⋅⟩ represents the Kronecker pairing.
Proposition 6.30. Let 𝑀 be a connected compact Riemann surface.

ℤ.
(a) The degree map descends to a surjective homomorphism from Pic(𝑀) to

(b) Two holomorphic line bundles on 𝑀 are smoothly isomorphic if and only
they have the same degree.

Proof. The fact that 𝑐(𝐿) depends only on the isomorphism class of 𝐿 implies
immediately that the degree map descends to Pic(𝑀).
Other Sheaf Cohomology Theories 183

Because 𝑀 has complex dimension 1, there are no nonzero (0, 2)-forms, so the
Dolbeault group 𝐻 0,2 (𝑀) is zero. Thus 𝐻 2 (𝑀; 𝒪) = 0 by the Dolbeault theorem,
so (6.36) implies that the map 𝛿∗ ∶ 𝐻 2 (𝑀; 𝒪 ∗ ) → ∶ 𝐻 2 (𝑀; ℤ) is surjective, and
thus so is the Chern class map 𝑐 ∶ Pic(𝑀) → 𝐻Sing 2
(𝑀; ℤ). Because 𝑀 is con-

[LeeTM, Thms. 6.15 & 10.22] shows that it is homeomorphic either to 𝕊2 or to

nected, compact, and orientable, the classification theorem for compact surfaces

a connected sum of one or more copies of the torus 𝕊1 × 𝕊1 . Let 𝑔 be the genus
of 𝑀, which is defined to be 0 if 𝑀 is homeomorphic to 𝕊2 and to be 𝑔 if it is
homeomorphic to a connected sum of 𝑔 ≥ 1 tori. The first singular homology
group of 𝑀 is isomorphic to the free abelian group ℤ2𝑔 [LeeTM, Cor. 13.15], and

map 𝜅 ∶ 𝐻Sing
2
(𝑀; ℤ) → Hom(𝐻2 (𝑀); ℤ) is an isomorphism. On the other hand,
therefore the universal coefficient theorem (Prop. 6.16) implies that the Kronecker

since 𝐻2 (𝑀) is an infinite cyclic group generated by the fundamental class [𝑀]
(Prop. 6.15), it follows that the evaluation map 𝐸 ∶ Hom(𝐻2 (𝑀); ℤ) → ℤ given
by 𝐸(𝜑) = 𝜑([𝑀]) is also an isomorphism.
The degree map can be written as the following composition:
𝑐 𝜅 𝐸
Pic(𝑀) ⟶ 𝐻Sing
2
(𝑀; ℤ) ⟶ Hom(𝐻2 (𝑀), ℤ) ⟶ ℤ.

Since 𝑐 is a surjective homomorphism and 𝜅 and 𝐸 are both isomorphisms, claim

(a) follows. Then (b) follows immediately from Theorem 6.29. □

Its kernel is the Picard variety Pic0 (𝑀) defined above.

The degree homomorphism is surjective, but in most cases it is not injective.

Later, we will develop some effective methods for computing the degrees of
line bundles on a Riemann surface.

Other Sheaf Cohomology Theories

The Čech construction of sheaf cohomology that we have described is convenient
for differential geometry because it makes the interconnections among holomorphic
line bundles, differential forms, and algebraic topology relatively transparent. But
as we noted earlier, its effectiveness is limited to paracompact Hausdorff spaces.
For some applications, this is an intolerable restriction. This is especially true in
algebraic geometry, because the Zariski topology is not Hausdorff.
There are other definitions of sheaf cohomology that are more general. Since
you might encounter those definitions in other books, for completeness we briefly
describe the most important ones here and show that they yield cohomology groups
that are isomorphic to the Čech groups when applied to paracompact Hausdorff
spaces. The material in the remainder of this chapter will not be used anywhere
else in the book, so you are free to ignore it if you wish.
184 6. Sheaf Cohomology

Most other approaches to sheaf cohomology groups are based on constructing

a certain acyclic resolution and defining the sheaf cohomology groups to be the
cohomology groups of the associated global section sequence. Of course, “acyclic”
does not make sense until the sheaf cohomology groups have been defined, so the
definitions proceed by defining a canonical resolution by a certain class of sheaves
that will turn out to be acyclic in the given theory.
There are two common approaches to defining sheaf cohomology in this way.

Roger Godement [God73], uses a resolution by flasque sheaves: A sheaf ℱ on a

(A less common third approach is described in Problem 6-7.) The first, due to

topological space 𝑀 is said to be flasque if for every open subset 𝑈 ⊆ 𝑀, the

restriction map 𝑟𝑀𝑈 ∶ ℱ (𝑀) → ℱ (𝑈 ) is surjective. It follows that every restriction
map 𝑟𝑉 ∶ ℱ (𝑈 ) → ℱ (𝑉 ) is surjective because 𝑟𝑈
𝑈
𝑉 ∘ 𝑟𝑈 = 𝑟𝑉 . (Some authors use
𝑀 𝑀

the term flabby, which is the English translation of the French word flasque.) Gode-
ment showed that every sheaf of abelian groups admits a canonical resolution by
flasque sheaves (the Godement resolution, described in Problem 6-6), and defined
the sheaf cohomology groups to be the cohomology groups of the associated global
section sequence of that resolution. The advantage of this definition is that short
exact sheaf sequences yield long exact cohomology sequences on any topological
space, not just a paracompact Hausdorff one.

a resolution by injective sheaves: a sheaf 𝒮 on 𝑀 is injective if whenever ℬ is a

The second common approach, due to Alexander Grothendieck [Gro57], uses

sheaf on 𝑀 and 𝒜 is a subsheaf of ℬ, every sheaf morphism 𝐹 ∶ 𝒜 → 𝒮 extends

to ℬ; that is, there is a sheaf morphism 𝐹̃ ∶ ℬ → 𝒮 making the following diagram
commute:

ℬO ❇
❇ 𝐹̃

𝒜 / 𝒮.
❇

𝐹
? ❇

Grothendieck showed that every sheaf of abelian groups admits a canonical resolu-
tion by injective sheaves, and defined sheaf cohomology groups as the cohomology
groups of the associated global section sequence. This is the most general defini-
tion of all, because it can be expressed in purely category-theoretic terms, and thus
makes sense for sheaves over objects even more general than topological spaces.
Both of these constructions lead to definitions of sheaf cohomology groups that
satisfy all the essential properties that we have proved for the Čech groups: they are
functorial; zero-degree cohomology is naturally isomorphic to the group of global
sections; and short exact sheaf sequences yield long exact cohomology sequences.
The details can be found, for example, in [Bre97, Chap. II]. In this section, we will
prove that both flasque sheaves and injective sheaves on a paracompact Hausdorff
Other Sheaf Cohomology Theories 185

space are acyclic in Čech cohomology, so it follows from the de Rham–Weil the-
orem that the cohomology groups defined by either the Godement construction or
the Grothendieck construction are isomorphic to the ones we have defined.
We begin by describing the relationship between these two types of sheaves.
Theorem 6.31. On any topological space, every injective sheaf of abelian groups
is flasque.

Proof. Let 𝒮 be an injective sheaf of abelian groups on a topological space 𝑀.

Suppose 𝑈 ⊆ 𝑀 is an open subset and 𝑠 ∈ 𝒮 (𝑈 ). Let ℤ be the constant sheaf on
𝑀 with coefficients in ℤ, and define a subsheaf ℤ𝑈 ⊆ ℤ by
ℤ𝑈 (𝑉 ) = {𝑘 ∈ ℤ(𝑉 ) ∶ 𝑘(𝑥) = 0 for 𝑥 ∈ 𝑉 ∖ 𝑈 }.
(Recall that elements of ℤ(𝑉 ) are locally constant functions from 𝑉 to ℤ.) Because
𝒮 is a sheaf of ℤ-modules (see Example 5.4(c)), we can use the section 𝑠 to define
a sheaf morphism 𝐼 ∶ ℤ𝑈 → 𝒮 as follows: given an open subset 𝑉 ⊆ 𝑀 and a
section 𝑘 ∈ ℤ𝑈 (𝑉 ), let 𝐼(𝑘) ∈ 𝒮 (𝑉 ) be the section satisfying
𝐼(𝑘)|𝑉 ∩𝑈 = (𝑘|𝑉 ∩𝑈 )(𝑠|𝑉 ∩𝑈 ),
𝐼(𝑘)|𝑉 ∩𝑘−1 (0) = 0.

The fact that 𝑘 is locally constant implies that 𝑘−1 (0) is open, so the gluing and
locality properties of 𝒮 ensure that 𝐼(𝑘) is uniquely defined. Because 𝒮 is injective,
𝐼 extends to a morphism 𝐼̃∶ ℤ → 𝒮 . Define a section 𝑠 ̃ ∈ 𝒮 (𝑀) by 𝑠 ̃ = 𝐼(1), ̃
where 1 ∈ ℤ(𝑀) is the constant function with value 1. It satisfies
̃ 𝑈 = 𝐼(1|
𝑠|̃ 𝑈 = 𝐼(1)| ̃ 𝑈 ) = 𝐼(1|𝑈 ) = 𝑠,

so 𝑟𝑀
𝑈 (𝑠)̃ = 𝑠, showing that 𝑟𝑈 is surjective.
𝑀
□

Problem 6-5 describes an example that shows the converse of this theorem is
not true.
Here are two essential properties of flasque sheaves.
Lemma 6.32 (Properties of Flasque Sheaves). Suppose ℛ, 𝒮 , and 𝒯 are sheaves
of abelian groups on a topological space 𝑀, and the following sheaf sequence is
exact:
𝐹 𝐺
(6.37) 0 → ℛ ⟶ 𝒮 ⟶ 𝒯 → 0.
(a) If ℛ is flasque, then for every open subset 𝑈 ⊆ 𝑀, the following sequence
of abelian groups is exact:
𝐹𝑈 𝐺𝑈
(6.38) 0 → ℛ(𝑈 ) ⟶ 𝒮 (𝑈 ) ⟶ 𝒯 (𝑈 ) → 0.
(b) If ℛ and 𝒮 are flasque, then so is 𝒯 .
186 6. Sheaf Cohomology

Proof. Suppose first that ℛ is flasque. We know from Proposition 5.24 that the
sequence (6.38) is exact at ℛ(𝑈 ) and 𝒮 (𝑈 ), so we need only prove that the homo-
morphism 𝐺𝑈 ∶ 𝒮 (𝑈 ) → 𝒯 (𝑈 ) is surjective.
Let 𝑡 ∈ 𝒯 (𝑈 ) be arbitrary. The fact that the sheaf sequence (6.37) is exact
means that for each 𝑝 ∈ 𝑈 , there is a neighborhood 𝑉 of 𝑝 contained in 𝑈 and
a section 𝑠 ∈ 𝒮 (𝑉 ) such that 𝐺𝑉 (𝑠) = 𝑡|𝑉 . Let 𝑃 be the set of all pairs of the
form (𝑉 , 𝑠) where 𝑉 is an open subset of 𝑈 and 𝑠 ∈ 𝒮 (𝑉 ) satisfies 𝐺𝑉 (𝑠) = 𝑡|𝑉 .
Give 𝑃 a partial order by saying that (𝑉 , 𝑠) ≤ (𝑉 ′ , 𝑠′ ) if 𝑉 ⊆ 𝑉 ′ and 𝑠′ |𝑉 = 𝑠. If
𝑄 = {(𝑉𝛼 , 𝑠𝛼 ) ∶ 𝛼 ∈ 𝐴} is any totally ordered subset of 𝑃 , then the pair (𝑉∞ , 𝑠∞ ) is
an upper bound for 𝑄, where 𝑉∞ = ⋃𝛼 𝑉𝛼 and 𝑠∞ ∈ 𝒮 (𝑉∞ ) is defined as follows:
the fact that 𝑄 is totally ordered guarantees that for any 𝛼, 𝛽 ∈ 𝐴, either 𝑉𝛼 ⊆ 𝑉𝛽
or vice versa, and in either case 𝑠𝛼 |𝑉𝛼 ∩𝑉𝛽 = 𝑠𝛽 |𝑉𝛼 ∩𝑉𝛽 , so by the gluing property
there exists 𝑠∞ ∈ 𝒮 (𝑉∞ ) whose restriction to each 𝑉𝛼 is equal to 𝑠𝛼 . It follows that
𝐺𝑉∞ (𝑠∞ ) = 𝑡|𝑉∞ , so (𝑉∞ , 𝑠∞ ) ∈ 𝑃 . By Zorn’s lemma, there is a maximal element
(𝑉 , 𝑠) ∈ 𝑃 .
We will show that 𝑉 = 𝑈 , which means that 𝐺𝑈 (𝑠) = 𝑡, proving that 𝐺𝑈 is
surjective. Assume for contradiction that 𝑉 ≠ 𝑈 , and let 𝑝 be a point of 𝑈 ∖ 𝑉 .
Then there exist a neighborhood 𝑊 of 𝑝 in 𝑈 and a section 𝜎 ∈ 𝒮 (𝑊 ) such that
𝐺𝑊 (𝜎) = 𝑡|𝑊 . Since 𝐺𝑉 ∩𝑊 (𝜎|𝑉 ∩𝑊 ) = 𝑡|𝑉 ∩𝑊 = 𝐺𝑉 ∩𝑊 (𝑠|𝑉 ∩𝑊 ), the fact that
0 → ℛ(𝑉 ∩ 𝑊 ) → 𝒮 (𝑉 ∩ 𝑊 ) → 𝒯 (𝑉 ∩ 𝑊 ) is exact implies there exists 𝑟 ∈
ℛ(𝑉 ∩ 𝑊 ) such that
(6.39) 𝑠|𝑉 ∩𝑊 − 𝜎|𝑉 ∩𝑊 = 𝐹𝑉 ∩𝑊 (𝑟).
Because ℛ is flasque, there is a section 𝑟 ̃ ∈ ℛ(𝑊 ) that restricts to 𝑟 on 𝑉 ∩ 𝑊 .
Let 𝜎
̃ = 𝜎 + 𝐹𝑊 (𝑟)̃ ∈ 𝒮 (𝑊 ). It follows from (6.39) that the restrictions of 𝑠 and 𝜎
̃
agree on 𝑉 ∩ 𝑊 , so there is a section 𝑠 ̃ ∈ 𝒮 (𝑉 ∪ 𝑊 ) that satisfies
𝑠|̃ 𝑉 = 𝑠, 𝑠|̃ 𝑊 = 𝜎
̃.
Then 𝐺𝑉 ∪𝑊 (𝑠)̃ agrees with 𝐺𝑉 (𝑠) on 𝑉 and with 𝐺𝑊 (̃ 𝜎 ) = 𝐺𝑊 (𝜎) on 𝑊 , so by
the locality property 𝐺𝑉 ∪𝑊 (𝑠)̃ = 𝑡|𝑉 ∪𝑊 , showing that (𝑉 ∪ 𝑊 , 𝑠)̃ ∈ 𝑃 . Since it
is strictly larger than (𝑉 , 𝑠), this contradicts the maximality of (𝑉 , 𝑠), thus proving

To prove (b), assume in addition that 𝒮 is flasque, and suppose 𝑈 ⊆ 𝑀 is open

(a).

and 𝑡 ∈ 𝒯 (𝑈 ). By the result of part (a), there is a section 𝑠 ∈ 𝒮 (𝑈 ) such that

𝐺𝑈 (𝑠) = 𝑡. Since 𝒮 is flasque, there is a section 𝑠 ̃ ∈ 𝒮 (𝑀) such that 𝑠|̃ 𝑈 = 𝑠, and
then 𝑡 ̃ = 𝐺𝑀 (𝑠)̃ ∈ 𝒯 (𝑀) satisfies
𝑡|̃ 𝑈 = 𝐺𝑀 (𝑠)|
̃ 𝑈 = 𝐺𝑈 (𝑠|̃ 𝑈 ) = 𝐺𝑈 (𝑠) = 𝑡,
showing that 𝒯 is flasque. □

sheaf 𝒮 , we define the sheaf of rough sections of 𝓢, denoted by 𝒮̂, by letting 𝒮̂(𝑈 )
The principal source of flasque sheaves is the following construction. For any
Other Sheaf Cohomology Theories 187

be the set of all functions 𝜎 ∶ 𝑈 → Et(𝒮 ) satisfying 𝜋 ∘ 𝜎 = Id𝑈 ; it is a sheaf by

the result of Exercise 5.1(c). If 𝒮 is a sheaf of abelian groups, then so is 𝒮̂ with
addition defined pointwise, because each stalk of Et(𝒮 ) inherits an abelian group
structure from 𝒮 . (In the literature, 𝒮̂ is often called the “sheaf of discontinuous
sections of 𝒮 ”; but of course not all of the sections in 𝒮̂(𝑈 ) are discontinuous.)
Lemma 6.33 (Properties of the Sheaf of Rough Sections). Suppose 𝒮 is a sheaf
of abelian groups on a topological space 𝑀, and 𝒮̂ is its sheaf of rough sections.
(a) 𝒮̂ is flasque.
(b) There is a canonical injective sheaf morphism 𝜄𝒮 ∶ 𝒮 ↪ 𝒮̂.
(c) Given a sheaf morphism 𝐹 ∶ 𝒮 → 𝒯 , there is a unique sheaf morphism
𝐹̂ ∶ 𝒮̂ → 𝒯̂ making the following diagram commute:
𝐹 /
𝒮 𝒯
𝜄𝒮 𝜄
 𝒯
𝐹̂ / ̂
(6.40)

𝒮̂ 𝒯.


(d) If 𝑀 is a paracompact Hausdorff space, then 𝒮̂ is fine.

Proof. Every section in 𝒮̂(𝑈 ) extends to a section in 𝒮̂(𝑀) by defining it to be

zero outside of 𝑈 , so 𝒮̂ is flasque.
The morphism 𝜄𝒮 ∶ 𝒮 → 𝒮̂ is defined by sending 𝑠 ∈ 𝒮 (𝑈 ) to the (continuous)
section 𝑠+ ∈ 𝒮̂(𝑈 ) defined by (5.6). Since two elements of 𝒮 (𝑈 ) that have the same
germ at every point are equal by the locality property, 𝜄𝒮 is injective.
If 𝐹 ∶ 𝒮 → 𝒯 is a sheaf morphism, the morphism 𝐹̂ ∶ 𝒮̂ → 𝒯̂ is defined by
𝐹̂𝑈 (𝜎)(𝑝) = 𝐹𝑝 (𝜎(𝑝))
for 𝜎 ∈ 𝒮̂(𝑈 ), where 𝐹𝑝 is the stalk homomorphism determined by 𝐹 . Since 𝐹̂𝑈 is

and 𝐹̂ respects the group structure on stalks because each 𝐹𝑝 does. For any section
defined pointwise, it commutes with restrictions and thus defines a sheaf morphism;

𝑠 ∈ 𝒮 (𝑈 ),
𝐹̂𝑈 ∘ 𝜄𝒮 (𝑠)(𝑝) = 𝐹𝑝 (𝑠+ (𝑝)) = 𝐹𝑝 ([𝑠]𝑝 ) = [𝐹𝑈 (𝑠)]𝑝
= (𝐹𝑈 (𝑠))+ (𝑝) = 𝜄𝒯 ∘ 𝐹𝑈 (𝑠)(𝑝),
showing that (6.40) commutes. For every section 𝜎 ∈ 𝒮̂(𝑈 ), the value 𝜎(𝑝) at each
𝑝 ∈ 𝑈 is the germ at 𝑝 of some section 𝑠 ∈ 𝒮 (𝑉 ) on some neighborhood 𝑉 of 𝑝,

𝐹̂𝑈 (𝜎)(𝑝) = 𝐹𝑝 ([𝑠]𝑝 ) = [𝐹𝑉 (𝑠)]𝑝 .

and

This shows that 𝐹̂ is completely determined by 𝐹 , thus proving uniqueness.

Now suppose 𝑀 is a paracompact Hausdorff space. To see that 𝒮̂ is fine, sup-
pose {𝑈𝛽 }𝛽∈𝐵 is a locally finite open cover of 𝑀, and let {𝜓𝛽 }𝛽∈𝐵 be a subordinate
188 6. Sheaf Cohomology

topological partition of unity. Choose a well-ordering of the index set 𝐵, and for
each 𝛽 ∈ 𝐵 define a morphism 𝜂𝛽 ∶ 𝒮̂ → 𝒮̂ on the stalk level by

𝜎(𝑥) if 𝛽 is the smallest index for which 𝜓𝛽 (𝑥) > 0,

(𝜂𝛽 𝜎)(𝑥) =
{0 otherwise.
The support of 𝜂𝛽 is contained in that of 𝜓𝛽 and therefore in 𝑈𝛽 , and it is straight-
forward to check that the collection {𝜂𝛽 }𝛽∈𝐵 is a sheaf partition of unity. □

For convenience, we will indulge in a slight abuse of notation and consider 𝒮

to be a subsheaf of 𝒮̂ by identifying it with the image of the morphism 𝜄𝒮 , which
is exactly the sheafification 𝒮 + .
Theorem 6.34. Let 𝑀 be a paracompact Hausdorff space. Every flasque sheaf of
abelian groups on 𝑀 is acyclic in Čech cohomology.

Proof. We will prove by induction on 𝑞 that for every 𝑞 ≥ 1 and every flasque sheaf
𝒮 on 𝑀, we have 𝐻 𝑞 (𝑀; 𝒮 ) = 0.
Suppose 𝒮 is a flasque sheaf on 𝑀, and let 𝒮̂ be its sheaf of rough sections.
We have an exact sheaf sequence
(6.41) 0 → 𝒮 ↪ 𝒮̂ → 𝒮̂/𝒮 → 0.
The corresponding long exact sequence in Čech cohomology contains the segment
(6.42) 𝐻 0 (𝑀; 𝒮̂) → 𝐻 0 (𝑀; 𝒮̂/𝒮 ) → 𝐻 1 (𝑀; 𝒮 ) → 𝐻 1 (𝑀; 𝒮̂).
Since 𝒮̂ is fine, the rightmost group in (6.42) is trivial. By Theorem 6.7, the first
map is equivalent to the global section map 𝒮̂(𝑀) → (𝒮̂/𝒮 )(𝑀), which is surjec-
tive by Lemma 6.32; thus 𝐻 1 (𝑀; 𝒮 ) = 0 for every flasque sheaf 𝒮 .
Now let 𝑞 ≥ 1, and assume we have proved that 𝐻 𝑞 (𝑀; 𝒮 ) = 0 for every
flasque sheaf 𝒮 . Suppose 𝒮 is flasque, and consider the following portion of the
long exact sequence associated with (6.41):
𝐻 𝑞 (𝑀; 𝒮̂) → 𝐻 𝑞 (𝑀; 𝒮̂/𝒮 ) → 𝐻 𝑞+1 (𝑀; 𝒮 ) → 𝐻 𝑞+1 (𝑀; 𝒮̂).
The first and last groups above are zero because 𝒮̂ is fine. It follows from Lemma
6.32 that 𝒮̂/𝒮 is flasque, so the inductive hypothesis implies 𝐻 𝑞+1 (𝑀; 𝒮 ) ≅
𝐻 𝑞 (𝑀; 𝒮̂/𝒮 ) = 0, thus completing the induction. □
Corollary 6.35. On a paracompact Hausdorff space, the sheaf cohomology groups
defined by either the Godement construction or the Grothendieck construction are
isomorphic to those defined by the Čech construction.

Proof. Either construction starts with a resolution by sheaves that are acyclic in
Čech cohomology by Theorems 6.31 and 6.34, so the result follows from the de
Rham–Weil theorem. □
Problems 189

Problems
6-1. Suppose 𝑀 is a positive-dimensional complex manifold. Show that the
sheaf 𝒪 of holomorphic functions on 𝑀 and the constant sheaf ℂ on 𝑀
are not fine. [Hint: Let 𝑈 , 𝑉 be proper open subsets of 𝑀 whose union
is 𝑀 and whose intersection is nonempty, and show that there is no sheaf
partition of unity subordinate to the cover 𝒰 = {𝑈 , 𝑉 }.]
6-2. Let 𝑀 be a paracompact Hausdorff space and ℛ a fine sheaf of commu-
tative rings on 𝑀. Show that every sheaf of ℛ-modules on 𝑀 is fine.
6-3. Let 𝑀 be a smooth manifold, and let 𝒰 = {𝑈𝛼 }𝛼∈𝐴 be an indexed open
cover of 𝑀 such that each nonempty finite intersection 𝑈𝛼0 ∩ ⋯ ∩ 𝑈𝛼𝑝 is

ian metric on 𝑀 and choosing a geodesically convex neighborhood of

contractible. (Such a cover can be constructed by choosing a Riemann-

each point [LeeRM, Thm. 6.17], and noting that intersections of geodesi-
cally convex sets are geodesically convex and therefore contractible.) By

the de Rham–Weil isomorphisms ℛ1 ∶ 𝐻dR 1

(𝑀; ℂ) → 𝐻 1 (𝑀; ℂ) and
following through the proof of the de Rham–Weil theorem, show that

ℛ2 ∶ 𝐻dR2
(𝑀; ℂ) → 𝐻 2 (𝑀; ℂ) can be described as follows.
(a) Let 𝜂 be a closed 1-form on 𝑀. For each 𝛼, there is a smooth function
𝑢𝛼 on 𝑈𝛼 such that 𝜂 |𝑈 = 𝑑𝑢𝛼 . Then
𝛼

𝑎𝛼𝛽 = 𝑢𝛽 |𝑈 − 𝑢 𝛼 |𝑈
𝛼 ∩𝑈𝛽 𝛼 ∩𝑈𝛽

defines a 1-cocycle 𝑎 on 𝒰 with coefficients in ℂ, and ℛ1 [𝜂] = [[𝑎]].

(b) Let 𝜂 be a closed 2-form on 𝑀. For each 𝛼, there is a smooth 1-
form 𝜑𝛼 on 𝑈𝛼 such that 𝜂|𝑈𝛼 = 𝑑𝜑𝛼 ; and for each 𝛼 and 𝛽 such that
𝑈𝛼 ∩ 𝑈𝛽 ≠ ∅, there is a smooth function 𝑢𝛼𝛽 on 𝑈𝛼 ∩ 𝑈𝛽 such that
𝜑𝛽 |𝑈 ∩𝑈 − 𝜑𝛼 |𝑈 ∩𝑈 = 𝑑𝑢𝛼𝛽 . Then
𝛼 𝛽 𝛼 𝛽

𝑎𝛼𝛽𝛾 = (𝑢𝛽𝛾 − 𝑢𝛼𝛾 + 𝑢𝛼𝛽 )|𝑈 ∩𝑈 ∩𝑈

𝛼 𝛽 𝛾

defines a 2-cocycle on 𝒰 with coefficients in ℂ, and ℛ2 [𝜂] = [[𝑎]].

6-4. Let 𝑀 be a complex manifold. Let 𝒫 denote the sheaf of pluriharmonic
functions on 𝑀 (see Problem 4-7), and ℰℝ the sheaf of smooth real-valued
functions. For each 𝑞 ≥ 1, let ℱ 𝑞 denote the sheaf of smooth real (𝑞 + 1)-
forms whose (𝑞 + 1, 0) and (0, 𝑞 + 1)-parts are zero; in other words, ℱ 𝑞
is the sheaf of smooth real-valued forms taking values in Λ𝑞,1 𝑀 ⊕ ⋯ ⊕
Λ1,𝑞 𝑀. Show that the following sheaf sequence is a fine resolution of 𝒫 :

𝑖𝜕𝜕 𝑑 𝑑 𝑑 𝑑
0 → 𝒫 ↪ ℰℝ ⟶ ℱ 1 → ℱ 2 → ⋯ → ℱ 𝑞 → ⋯ .
190 6. Sheaf Cohomology

Conclude that for 𝑞 ≥ 2,

Ker (𝑑 ∶ ℱ 𝑞 (𝑀) → ℱ 𝑞+1 (𝑀))
𝐻 𝑞 (𝑀; 𝒫 ) ≅ .
Im (𝑑 ∶ ℱ 𝑞−1 (𝑀) → ℱ 𝑞 (𝑀))
State the analogous result for 𝑞 = 1. [Hint: For the proof of exactness at
ℱ 𝑞 , if 𝛽 is a local section of ℱ 𝑞 and 𝛽 = 𝑑𝛼, write 𝛼 = 𝛼 (𝑞,0) + 𝛼
̃ + 𝛼 (0,𝑞)
with 𝛼 ̃ a section of ℱ 𝑞−1
, and show that locally 𝑑𝛼 (𝑞,0)
= 𝑑𝜕𝜎 for some
(𝑞 − 1, 0) form 𝜎.]
6-5. Let 𝑀 be any nonempty topological space, and let ℤ𝑝 and ℝ𝑝 be the
skyscraper sheaves at a point 𝑝 ∈ 𝑀 with values in ℤ and ℝ, respec-
tively, so ℤ𝑝 is a subsheaf of ℝ𝑝 . Show that ℤ𝑝 is flasque but not injec-
tive, by showing that the identity morphism ℤ𝑝 → ℤ𝑝 does not extend to
a morphism ℝ𝑝 → ℤ𝑝 .

sheaf. Let 𝒮 be a sheaf of abelian groups on a topological space 𝑀.

6-6. This problem describes how to construct the Godement resolution of a

Define 𝒮 0 = 𝒮̂ (the sheaf of rough sections of 𝒮 ). Next, let 𝒬 1 = 𝒮 0 /𝒮 ,

where we identify 𝒮 with a subsheaf of 𝒮̂ via the canonical injection 𝜄𝒮
given by Lemma 6.33; and let 𝒮 1 = 𝒬 ̂1 . Let 𝑑 ∶ 𝒮 0 → 𝒮 1 be the
0

𝒬1 ̂
composition
Π 𝜄
𝒮 0 −→ 𝒬 1 −−→ 𝒬 1,

6.33. Then by induction define 𝒬 𝑗+1 = 𝒬 ̂

̂𝑗 /𝒬 𝑗 and 𝒮 𝑗+1 = 𝒬
where the first map is given by Proposition 5.22 and the second by Lemma
𝑗+1 , with the

morphism 𝑑𝑗 ∶ 𝒮 → 𝒮
𝑗 𝑗+1

of 𝓢 is the following sequence of sheaf morphisms

defined as above. The Godement resolution

𝜄𝒮 𝑑0 𝑑1
0 → 𝒮 −−→ 𝒮 0 −−→ 𝒮 1 −−→ 𝒮 2 → ⋯ .
Prove that this is a flasque resolution.
6-7. Another approach that is sometimes used to define sheaf cohomology

defined as follows. If 𝒮 is a sheaf of abelian groups on a space 𝑀 and

groups (e.g., in [Wel08, Chap. 2]) is based on resolutions by soft sheaves,

𝐾 ⊆ 𝑀 is a closed subset, let 𝒮 (𝐾) denote the group of continuous sec-

tions of the étalé space Et(𝒮 ) over 𝐾. (This can be naturally identified
with the inverse image sheaf 𝑖−1 (𝒮 ), where 𝑖 ∶ 𝐾 → 𝑀 is inclusion. See
Problem 5-13). We define a restriction operator 𝑟𝑀 𝐾 ∶ 𝒮 (𝑀) → 𝒮 (𝐾) by
letting 𝑟𝐾 (𝑠) be the restriction to 𝐾 of the continuous function 𝑠+ ∶ 𝑀 →
𝑀

Et(𝒮 ) defined by (5.6). The sheaf 𝒮 is said to be soft if for every closed
subset 𝐾 ⊆ 𝑀, the restriction map 𝑟𝑀 𝐾 is surjective.
(a) Let 𝑀 be a paracompact Hausdorff space. Prove that both conclu-

(b) For every sheaf 𝒮 , prove that the sheaf of rough sections 𝒮̂ is soft.
sions of Lemma 6.32 hold if “flasque” is replaced by “soft.”
Problems 191

(c) Prove that every soft sheaf on a paracompact Hausdorff space is

(d) Let 𝒮 be a sheaf of abelian groups on a paracompact Hausdorff space,

acyclic in Čech cohomology.

and suppose we are given a resolution of 𝒮 by soft sheaves. Prove that

are isomorphic to the Čech cohomology groups of 𝒮 .

the cohomology groups of the corresponding global section sequence

6-9. Let 𝒮 be a sheaf of abelian groups on a paracompact Hausdorff space 𝑀.

6-8. Prove that every fine sheaf on a paracompact Hausdorff space is soft.

(a) 𝒮 is fine.
Show that the following are equivalent:

(b) The sheaf ℋom(𝒮 , 𝒮 ) is soft (see Problems 5-7 and 6-7).
(c) For each pair of disjoint closed subsets 𝐾, 𝐿 ⊆ 𝑀, there is a sheaf
morphism 𝐹 ∶ 𝒮 → 𝒮 that restricts to the identity on a neighborhood
of 𝐾 and to the zero morphism on a neighborhood of 𝐿.
[Hint: For (b) ⇒ (a), let {𝑈𝛼 }𝛼∈𝐴 be a locally finite open cover of 𝑀, and
choose another open cover {𝑉𝛼 }𝛼∈𝐴 satisfying 𝑉𝛼 ⊆ 𝑉 𝛼 ⊆ 𝑈𝛼 . Let Ψ be
the set of collections {𝜓𝛽 }𝛽∈𝐵 where 𝐵 ⊆ 𝐴, each 𝜓𝛽 is a sheaf morphism
𝒮 → 𝒮 satisfying supp 𝜓𝛽 ⊆ 𝑈𝛽 , and ∑𝛽∈𝐵 𝜓𝛽 restricts to the identity
on ⋃𝛽∈𝐵 𝑉 𝛽 . Give Ψ a partial order by saying {𝜓𝛽 }𝛽∈𝐵 ≤ {𝜓𝛽′ }𝛽∈𝐵 ′ if
𝐵 ⊆ 𝐵 ′ and 𝜓𝛽 = 𝜓𝛽′ for 𝛽 ∈ 𝐵, and use Zorn’s lemma.]

[Mun84, Chap. 5] or [Hat02, Chap. 3].) Suppose 𝐾 is a simplicial com-

6-10. This problem requires some knowledge of simplicial cohomology. (See

plex and |𝐾| is its underlying topological space. For each vertex 𝑣 of 𝐾,
the star of 𝒗, denoted by St 𝑣, is the union of the interiors of all simplices
that have 𝑣 as a vertex. If 𝑣0 , … , 𝑣𝑝 are the vertices of a simplex 𝜎 of 𝐾,
then St 𝑣0 ∩ ⋯ ∩ St 𝑣𝑝 is a neighborhood of 𝜎 in |𝐾|. Let 𝐺 be an abelian
group and let 𝒰 be the open cover of |𝐾| consisting of the stars of all the

(a) Show that 𝐻 𝑝 (𝒰; 𝐺) is isomorphic to the simplicial cohomology

vertices.

group 𝐻 𝑝 (𝐾; 𝐺).

(b) If 𝒱 is a refinement of 𝒰, show that there is a subdivision of 𝐾 whose
covering by open stars provides a further refinement of 𝒱 , and then

provide another proof that 𝐻Sing (𝑀; 𝐺) ≅ 𝐻 𝑝 (𝑀; 𝐺) for a triangu-

𝑝
use the isomorphism between singular and simplicial cohomology to

lable topological space 𝑀.

 10.1090/gsm/244/07

Chapter 7

Connections

Among the main tools in Riemannian geometry are connections, or covariant differ-
entiation operators, on vector bundles. Connections play an equally important role
in complex geometry, and this chapter is dedicated to exploring their fundamental
properties.
We begin with the definition of a connection on a smooth complex vector bun-
dle, and show how to construct connections and how to do computations with them.
Next we describe the curvature of a connection, which can be thought of as an ob-
struction to the existence of parallel local frames. We then show how a connection
can be used to construct a cohomology class associated with every smooth complex
vector bundle, called the first real Chern class; for line bundles, it is closely related
to the sheaf-theoretic Chern class defined in Chapter 3.
The last part of the chapter focuses on holomorphic vector bundles. If we en-
dow such a bundle with a Hermitian fiber metric, there are many connections on the
bundle that are compatible with the metric. But for holomorphic Hermitian vector
bundles, there is an additional condition, called compatibility with the holomorphic
structure, that allows us to single out a unique connection, called the Chern con-
nection. Here we introduce the Chern connection and study some of its properties.

Connections on Complex Vector Bundles

Let 𝐸 be a smooth complex vector bundle of rank 𝑚 on a smooth manifold 𝑀. A
connection on 𝐸 is a map ∇ ∶ Γ(𝑇 𝑀) × Γ(𝐸) → Γ(𝐸), written (𝑋, 𝜎) ↦ ∇𝑋 𝜎,
that is linear over 𝐶 ∞ (𝑀) in 𝑋, linear over ℂ in 𝜎, and satisfies the product rule
∇𝑋 (𝑓 𝜎) = 𝑓 ∇𝑋 𝜎 + (𝑋𝑓 )𝜎 for every smooth complex-valued function 𝑓 . The
expression ∇𝑋 𝜎 is called the covariant derivative of 𝝈 in the direction 𝑿. The
value of ∇𝑋 𝜎 at a point 𝑝 ∈ 𝑀 depends only on the value of 𝑋 at 𝑝 and the value

193
194 7. Connections

of 𝜎 in an arbitrarily small neighborhood of 𝑝 (see [LeeRM, Lemma 4.5], where this

Consequently any connection ∇ on 𝐸 determines a connection, still denoted by ∇,

is proved for real vector bundles; the proof for complex bundles is just the same).

on the restriction of 𝐸 to any open subset of 𝑀 [LeeRM, Prop. 4.3]. We will often
make use of this fact without further comment.
In addition to determining covariant derivatives with respect to specific vector

section 𝜎: this is the section ∇𝜎 of the bundle 𝑇 ∗ 𝑀 ⊗ 𝐸, which is canonically

fields, a connection also determines a total covariant derivative of each smooth

isomorphic to Hom(𝑇 𝑀, 𝐸), defined by (∇𝜎)(𝑋) = ∇𝑋 𝜎.

as well as real ones: if ∇ is a connection on 𝐸 and 𝑍 = 𝑋 +𝑖𝑌 is a smooth complex

We extend our connections by complex linearity to accept complex vector fields

vector field on 𝑀, we set

∇𝑍 𝜎 = ∇𝑋 𝜎 + 𝑖∇𝑌 𝜎.

This operation is easily seen to be linear over 𝐶 ∞ (𝑀; ℂ) in 𝑍, and it allows us

to regard the total covariant derivative of 𝜎 (still denoted by ∇𝜎) as a section of
𝑇ℂ∗ 𝑀 ⊗ 𝐸 ≅ Hom(𝑇ℂ 𝑀, 𝐸).

tions related to them in terms of local frames. Thus suppose (𝑠𝑗 ) = (𝑠1 , … , 𝑠𝑚 ) is
Because connections are determined locally, we can carry out most computa-

a smooth local frame for 𝐸 over a subset 𝑈 ⊆ 𝑀. For a given vector field 𝑋 on 𝑈 ,
we can express each of the covariant derivatives ∇𝑋 (𝑠𝑗 ) in terms of the same frame
(using the summation convention) as

(7.1) ∇𝑋 𝑠𝑗 = 𝜃𝑗𝑘 (𝑋)𝑠𝑘

for some smooth coefficients 𝜃𝑗𝑘 (𝑋). Formula (7.1) determines the covariant deriv-
ative of an arbitrary section 𝜎 = 𝜎 𝑗 𝑠𝑗 by

(7.2) ∇𝑋 (𝜎 𝑗 𝑠𝑗 ) = (𝑋𝜎 𝑗 )𝑠𝑗 + 𝜎 𝑗 𝜃𝑗𝑘 (𝑋)𝑠𝑘 .

Because ∇ is linear over 𝐶 ∞ (𝑀) in 𝑋, the coefficients 𝜃𝑗𝑘 define a matrix of

complex 1-forms, called the connection forms with respect to this frame. They
are smooth because their action on any smooth vector field 𝑋 can be written as
𝜃𝑗𝑘 (𝑋) = 𝜀𝑘 (∇𝑋 𝑠𝑗 ), where (𝜀𝑘 ) is the frame for 𝐸 ∗ dual to (𝑠𝑗 ). In terms of these
forms, we can write the total covariant derivative as

(7.3) ∇𝑠𝑗 = 𝜃𝑗𝑘 ⊗ 𝑠𝑘 ,

(7.4) ∇(𝜎 𝑗 𝑠𝑗 ) = 𝑑𝜎 𝑗 ⊗ 𝑠𝑗 + 𝜎 𝑗 𝜃𝑗𝑘 ⊗ 𝑠𝑘 .

Conversely, given an arbitrary matrix of smooth complex 1-forms 𝜃𝑗𝑘 on the do-
main 𝑈 of a smooth local frame for 𝐸, formula (7.2) determines a connection on 𝐸
over 𝑈 .
Connections on Complex Vector Bundles 195

If we have another local frame (𝑠𝑘̃ ), then where they overlap we can write

𝑠𝑘̃ = 𝜏𝑘 𝑠𝑗 ,
𝑗

for a GL(𝑚, ℂ)-valued transition function 𝜏 = (𝜏𝑘 ). To see how the connection
𝑗

forms change, we compute

∇𝑠𝑘̃ = 𝑑𝜏𝑘 ⊗ 𝑠𝑗 + 𝜏𝑘 𝜃𝑗𝑙 ⊗ 𝑠𝑙 = 𝑑𝜏𝑘 ⊗ (𝜏 −1 )𝑗 𝑠𝑝̃ + 𝜏𝑘 𝜃𝑗𝑙 ⊗ (𝜏 −1 )𝑙 𝑠𝑝̃ .

𝑗 𝑗 𝑗 𝑝 𝑗 𝑝

Comparing this with ∇𝑠𝑘̃ = 𝜃 𝑘̃ ⊗ 𝑠𝑝̃ , we find that

𝑝

𝜃 𝑝𝑘̃ = (𝜏 −1 )𝑗 𝑑𝜏𝑘 + (𝜏 −1 )𝑙 𝜃𝑗𝑙 𝜏𝑘 ,

𝑝 𝑗 𝑝 𝑗

or in matrix notation,

(7.5) 𝜃 ̃ = 𝜏 −1 𝑑𝜏 + 𝜏 −1 𝜃𝜏.

(Note that the 𝑑𝜏 and 𝜃 factors above are matrices of complex 1-forms, while the
other factors on the right-hand side are matrices of complex-valued functions. It
is important to observe the order of factors because matrix multiplication does not

Now suppose 𝐸 is endowed with a Hermitian fiber metric. A connection ∇ on

commute in general.)

𝐸 is compatible with the metric, or a metric connection, if the following identity

holds for all 𝑋 ∈ Γ(𝑇 𝑀) and 𝜎, 𝜏 ∈ Γ(𝐸):

(7.6) 𝑋⟨𝜎, 𝜏⟩ = ⟨∇𝑋 𝜎, 𝜏⟩ + ⟨𝜎, ∇𝑋 𝜏⟩.

When we apply this to a complex vector field 𝑍, the equation for compatibility with
the metric reads

(7.7) 𝑍⟨𝜎, 𝜏⟩ = ⟨∇𝑍 𝜎, 𝜏⟩ + ⟨𝜎, ∇𝑍 𝜏⟩,

because of the conjugate linearity of the Hermitian inner product in its second ar-
gument.
The next proposition gives an important property of the connection forms for a
metric connection.
Proposition 7.1. Suppose 𝐸 → 𝑀 is a smooth complex vector bundle with a Her-
mitian fiber metric and ∇ is a metric connection on 𝐸. The matrix of connection
1-forms with respect to any local orthonormal frame is skew-Hermitian:

𝜃𝑗𝑘 = −𝜃𝑘 .
𝑗

Proof. Let (𝑠𝑗 ) be a local orthonormal frame for 𝐸, and let 𝜃𝑗𝑘 be the corresponding
connection 1-forms. For every local complex vector field 𝑍, compatibility with the
196 7. Connections

metric implies
0 = 𝑍(𝛿𝑗𝑘 ) = 𝑍⟨𝑠𝑗 , 𝑠𝑘 ⟩ = ⟨∇𝑍 𝑠𝑗 , 𝑠𝑘 ⟩ + ⟨𝑠𝑗 , ∇𝑍 𝑠𝑘 ⟩

= ⟨𝜃𝑗𝑙 (𝑍)𝑠𝑙 , 𝑠𝑘 ⟩ + ⟨𝑠𝑗 , 𝜃𝑘𝑙 (𝑍)𝑠𝑙 ⟩ = 𝜃𝑗𝑙 (𝑍)𝛿𝑙𝑘 + 𝜃𝑘𝑙 (𝑍)𝛿𝑗𝑙

= 𝜃𝑗𝑘 (𝑍) + 𝜃𝑘 (𝑍).

𝑗
□

example, we can choose an open cover of 𝑀 by domains 𝑈𝛼 on each of which there

There are always plenty of connections compatible with a given metric: For

exists a local orthonormal frame for 𝐸, together with a smooth partition of unity
{𝜓𝛼 } subordinate to this cover; then define a connection ∇𝛼 on 𝐸|𝑈𝛼 by setting all
of the connection 1-forms to be zero, and define a global connection ∇ on 𝐸 by
blending these together with the partition of unity:
∇𝑋 𝜎 = 𝜓𝛼 ∇𝛼𝑋 𝜎.
∑
(7.8)
𝛼

► Exercise 7.2. Prove that (7.8) defines a metric connection on 𝐸.

The following lemma is a useful technical result about metric connections that
can simplify some computations. We will use it in Chapter 10.
Lemma 7.3. Suppose 𝐸 → 𝑀 is a smooth Hermitian vector bundle and ∇ is a
metric connection on 𝐸. In a neighborhood of each 𝑥0 ∈ 𝑀, there is a smooth
orthonormal frame (𝑠𝑗 ) for 𝐸 that satisfies ∇𝑠𝑗 = 0 at 𝑥0 .

Proof. Start with any orthonormal frame (𝑠𝑗 ), and let 𝜃𝑗𝑘 be the corresponding
connection forms. Proposition 7.1 shows that the matrix (𝜃𝑗𝑘 ) is skew-Hermitian.
Choose any smooth coordinates (𝑥𝑎 ) centered at 𝑥0 on an open set 𝑈 ⊆ 𝑀, and

𝜃𝑗𝑘 |𝑥 = 𝐴𝑘𝑗𝑎 𝑑𝑥𝑎 |𝑥

write
0 0

Let M(𝑟, ℂ) denote the space of all complex 𝑟 × 𝑟 matrices

𝐴𝑘𝑗𝑎 .
(where 𝑟 is the rank of 𝐸), and let 𝐹 ∶ 𝑈 → M(𝑟, ℂ) be the matrix-valued function
for some constants

𝐹𝑗𝑘 (𝑥) = 𝐴𝑘𝑗𝑎 𝑥𝑎 .

Then the fact that (𝜃𝑗𝑘 ) is skew-Hermitian implies that 𝐹 (𝑥) is skew-Hermitian for
all 𝑥. Define another matrix-valued function 𝐵(𝑥) = 𝑒−𝐹 (𝑥) ; it satisfies
𝐵𝑗𝑘 (𝑥0 ) = 𝛿𝑗𝑘 ; 𝜕𝑎 𝐵𝑗𝑘 (𝑥0 ) = −𝜕𝑎 𝐹𝑗𝑘 (𝑥0 ) = −𝐴𝑘𝑗𝑎 .

group, it follows that 𝐵(𝑥) is unitary for each 𝑥. Thus the local frame (𝑠𝑘̃ ) defined
Since the space of skew-Hermitian matrices is the Lie algebra of the unitary

𝑠𝑘̃ (𝑥) = 𝐵𝑘 (𝑥)𝑠𝑗 (𝑥)

𝑗
by
Connections on Complex Vector Bundles 197

is orthonormal. At 𝑥0 , 𝑠𝑘̃ = 𝑠𝑘 , and

∇𝜕𝑎 𝑠𝑘̃ = 𝜕𝑎 𝐵𝑘 (𝑥0 )𝑠𝑗 + 𝛿𝑘 𝐴𝑙𝑗𝑎 𝑠𝑙 = 0.

𝑗 𝑗

Thus (𝑠𝑘̃ ) satisfies the conclusion of the lemma. □

Covariant Derivatives Along Curves

A familiar construction in Riemannian geometry is to use a connection on the
tangent bundle to define a covariant derivative operator acting on vector fields along
curves (see [LeeRM, Thm. 4.24]). The same construction can be carried out for

Suppose 𝐸 → 𝑀 is a smooth complex vector bundle and 𝛾 ∶ 𝐼 → 𝑀 is a

arbitrary vector bundles; here we briefly describe how that works.

smooth curve (here 𝐼 ⊆ ℝ is some interval). A section of 𝑬 along 𝜸 is a continuous

map 𝜎 ∶ 𝐼 → 𝐸 satisfying 𝜎(𝑡) ∈ 𝐸𝛾(𝑡) for each 𝑡 ∈ 𝐼. A section along 𝛾 is said to
be extendible if there is a section 𝜎
̃ of 𝐸 on an open set containing the image of 𝛾
such that 𝜎(𝑡) = 𝜎
̃(𝛾(𝑡)) for each 𝑡 ∈ 𝐼.
Proposition 7.4 (Covariant Derivative Along a Curve). Suppose 𝐸 → 𝑀 is a
smooth complex vector bundle and ∇ is a connection on 𝐸. For each smooth curve
𝛾 ∶ 𝐼 → 𝑀, there is a unique operator 𝐷𝑡 that takes smooth sections of 𝐸 along 𝛾
to smooth sections along 𝛾, satisfying
(i) LINEARITY OVER ℝ: 𝐷𝑡 (𝑎𝜎 + 𝑏𝜏) = 𝑎𝐷𝑡 𝜎 + 𝑏𝐷𝑡 𝜏 for 𝑎, 𝑏 ∈ ℝ.
(ii) PRODUCT RULE: 𝐷𝑡 (𝑓 𝜎) = 𝑓 ′ 𝜎 + 𝑓 𝐷𝑡 𝜎 for 𝑓 ∈ 𝐶 ∞ (𝐼).
(iii) If 𝜎 is an extendible smooth section along 𝛾, then for every smooth exten-
sion 𝜎̃ of 𝜎, we have 𝐷𝑡 𝜎(𝑡) = ∇𝛾 ′ (𝑡) 𝜎
̃.

► Exercise 7.5. Show how to adapt the proof of [LeeRM, Thm. 4.24] to prove
this proposition.

For a smooth vector bundle 𝐸 → 𝑀 with a connection ∇, a smooth section 𝜎

of 𝐸 along a smooth curve 𝛾 ∶ 𝐼 → 𝑀 is said to be parallel along 𝜸 if 𝐷𝑡 𝜎 ≡ 0.
Proposition 7.6 (Existence and Uniqueness of Parallel Transport). Let 𝐸 → 𝑀
be a smooth vector bundle endowed with a connection ∇, and let 𝛾 ∶ 𝐼 → 𝑀 be a
smooth curve. Given 𝑡0 ∈ 𝐼 and an element 𝜎0 ∈ 𝐸𝛾(𝑡0 ) , there is a unique parallel
section 𝜎 along 𝛾 such that 𝜎(𝑡0 ) = 𝜎0 , called the parallel transport of 𝝈0 along 𝜸.

Proof. First assume the image of 𝛾 is contained in the domain of a smooth local
frame (𝑠1 , … , 𝑠𝑚 ) for 𝐸, and let 𝜃𝑗𝑘 be the connection 1-forms for this frame. We
can write 𝜎(𝑡) = 𝑓 𝑗 (𝑡)𝑠𝑗 (𝛾(𝑡)) (using the summation convention) for some smooth
functions 𝑓 1 , … , 𝑓 𝑚 ∶ 𝐼 → ℂ. Since the sections 𝑠𝑗 are extendible, Proposition
198 7. Connections

𝐷𝑡 𝜎(𝑡) = 𝑓 𝑗̇ (𝑡)𝑠𝑗 (𝛾(𝑡)) + 𝑓 𝑗 (𝑡)∇𝛾 ′ (𝑡) 𝑠𝑗

7.4 implies

= 𝑓 𝑗̇ (𝑡)𝑠𝑗 (𝛾(𝑡)) + 𝑓 𝑗 (𝑡)𝜃𝑗𝑘 (𝛾 ′ (𝑡))𝑠𝑘 (𝛾(𝑡))

= (𝑓 𝑘̇ (𝑡) + 𝑓 𝑗 (𝑡)𝜃𝑗𝑘 (𝛾 ′ (𝑡)))𝑠𝑘 (𝛾(𝑡))
(where dots represent derivatives with respect to 𝑡). Thus 𝜎 is parallel if and only
if its component functions satisfy the following system of linear ODEs on 𝐼:
𝑓 𝑘̇ (𝑡) = −𝑓 𝑗 (𝑡)𝜃 𝑘 (𝛾 ′ (𝑡)).
𝑗

of 𝐼 satisfying the initial conditions 𝑓 𝑗 (𝑡0 ) = 𝜎0 , where 𝜎0 = 𝜎0 𝑠𝑗 (𝛾(𝑡0 )).

𝑗 𝑗
It follows from [LeeRM, Thm. 4.31] that this system has a unique solution on all

For the general case, we can let 𝛽 be the supremum of all 𝑏 > 𝑡0 such that a
unique parallel transport 𝜎 exists on [0, 𝑏]. If 𝛽 < sup 𝐼, we can choose a local frame
on an open set containing 𝛾(𝛽 − 𝛿, 𝛽 + 𝛿) for some small 𝛿 > 0, and find a parallel
section 𝜎
̃ on that interval with initial value 𝜎̃(𝛽 − 𝛿/2) = 𝜎(𝛽 − 𝛿/2); by uniqueness,
𝜎
̃ agrees with 𝜎 on their common domain, so 𝜎 ̃ provides a parallel extension of 𝜎
past 𝛽, which is a contradiction. The same argument works for 𝑡 < 𝑡0 . □

A local section 𝜎 ∶ 𝑈 → 𝐸 defined on an open subset 𝑈 ⊆ 𝑀 is said to be

parallel if it is parallel along every smooth curve in 𝑈 . We end this section with
two properties of parallel transport whose proofs are left as exercises.
Proposition 7.7. If 𝐸 → 𝑀 is a smooth vector bundle endowed with a connection
∇, a local section 𝜎 ∶ 𝑈 → 𝐸 is parallel if and only if ∇𝜎 ≡ 0.
Proposition 7.8. Suppose 𝐸 → 𝑀 is a smooth Hermitian vector bundle and ∇ is
a metric connection on 𝐸.
(a) If 𝜎 and 𝜏 are parallel sections of 𝐸 along a smooth curve 𝛾, then ⟨𝜎, 𝜏⟩
is constant along 𝛾.
(b) If 𝜎 is a parallel section of 𝐸 (that is, ∇𝐸 ≡ 0) on a connected open subset
𝑈 ⊆ 𝑀, then |𝜎| is constant.

► Exercise 7.9. Prove the two preceding propositions.

Curvature
A fundamental local invariant of a Riemannian metric is the curvature of its Levi-
Civita connection. For connections on complex vector bundles, there is an analo-

Suppose 𝐸 → 𝑀 is a smooth complex vector bundle and ∇ is a connection on

gous local invariant, defined in essentially the same way.

𝐸. We define a map Θ ∶ Γ(𝑇ℂ 𝑀) × Γ(𝑇ℂ 𝑀) × Γ(𝐸) → Γ(𝐸), called the curvature

of ∇, by
Θ(𝑋, 𝑌 )𝜎 = ∇𝑋 ∇𝑌 𝜎 − ∇𝑌 ∇𝑋 𝜎 − ∇[𝑋,𝑌 ] 𝜎.
Curvature 199

The same argument as in the Riemannian case [LeeRM, Prop. 7.3] shows that Θ
is linear over 𝐶 ∞ (𝑀; ℂ) in all three arguments 𝑋, 𝑌 , 𝜎, and thus defines a smooth
section of the bundle 𝑇ℂ∗ 𝑀 ⊗𝑇ℂ∗ 𝑀 ⊗𝐸 ∗ ⊗𝐸. Moreover, Θ is clearly antisymmetric
in 𝑋 and 𝑌 , and 𝐸 ∗ ⊗𝐸 ≅ 𝐸 ⊗𝐸 ∗ is canonically isomorphic to the bundle End(𝐸);
so Θ can be viewed as an element of Γ(Λ2ℂ 𝑀 ⊗ End(𝐸)) = ℰ 2 (𝑀; End(𝐸)). A
connection ∇ is said to be flat if its curvature is identically zero.
To study the curvature more deeply, let us choose a smooth local frame (𝑠𝑗 )
for 𝐸, and let 𝜃𝑗𝑘 be the corresponding matrix of connection forms. To compute
the curvature, it suffices to compute its action on each basis section 𝑠𝑗 for arbitrary
𝑋, 𝑌 :
Θ(𝑋, 𝑌 )𝑠𝑗 = ∇𝑋 (𝜃𝑗𝑘 (𝑌 )𝑠𝑘 ) − ∇𝑌 (𝜃𝑗𝑘 (𝑋)𝑠𝑘 ) − 𝜃𝑗𝑘 ([𝑋, 𝑌 ])𝑠𝑘
= 𝑋 (𝜃𝑗𝑘 (𝑌 ))𝑠𝑘 + 𝜃𝑗𝑘 (𝑌 )𝜃𝑘𝑙 (𝑋)𝑠𝑙 − 𝑌 (𝜃𝑗𝑘 (𝑋))𝑠𝑘 − 𝜃𝑗𝑘 (𝑋)𝜃𝑘𝑙 (𝑌 )𝑠𝑙
− 𝜃𝑗𝑘 ([𝑋, 𝑌 ])𝑠𝑘
= (𝑑𝜃𝑗𝑙 (𝑋, 𝑌 ) + (𝜃𝑘𝑙 ∧ 𝜃𝑗𝑘 )(𝑋, 𝑌 ))𝑠𝑙 ,

of a 1-form [LeeSM, Prop. 14.29]. Thus Θ is represented locally by the matrix

where in the last line we have used the invariant formula for the exterior derivative

(Θ𝑗 ) of 2-forms, called the curvature forms of ∇ with respect to this frame, given
𝑙

by
(7.9) Θ𝑙𝑗 = 𝑑𝜃𝑗𝑙 + 𝜃𝑘𝑙 ∧ 𝜃𝑗𝑘 .
If we interpret 𝜃𝑗𝑙 and Θ𝑙𝑗 as the local expressions for End(𝐸)-valued forms, we can
use the wedge product of endomorphism-valued forms defined by (4.18) to write

Θ = 𝑑𝜃 + 𝜃 ∧ 𝜃.
(7.9) in the form

since 𝜃 is a matrix-valued 1-form, the wedge product 𝜃 ∧ 𝜃 need not be zero, as

There are a couple of caveats that need to be observed about this formula. First,

it would be if 𝜃 were a scalar-valued 1-form. Second, although Θ is a globally

defined End(𝐸)-valued 2-form as described above, we can only interpret 𝜃 as an

matrix representation for a global End(𝐸)-valued 1-form, then its transformation

endomorphism-valued form in the domain of a particular frame. If it were the local

law under a change of frames would be 𝜃 ̃ = 𝜏 −1 𝑑𝜏 instead of the more complicated

formula (7.5). Even so, the definition of wedge products involving 𝜃 works the
same as for a globally defined form.
The next proposition gives a qualitative interpretation of the curvature of a con-

(𝑠1 , … , 𝑠𝑘 ) for 𝐸 is said to be parallel if each of the sections 𝑠𝑗 is parallel.)

nection: it is the obstruction to the existence of parallel local frames. (A local frame

Proposition 7.10. Let 𝐸 → 𝑀 be a smooth complex vector bundle endowed with

a connection ∇. Then ∇ is flat if and only if every point of 𝑀 has a neighborhood
on which there exists a parallel local frame for 𝐸.
200 7. Connections

Proof. One direction is easy: if (𝑠1 , … , 𝑠𝑘 ) is a parallel local frame, then the con-
nection forms 𝜃𝑗𝑘 are all identically zero, so the curvature forms Θ𝑘𝑗 are zero as well.
Conversely, suppose Θ ≡ 0. Given 𝑝 ∈ 𝑀, we begin by showing that every
element 𝜎0 ∈ 𝐸𝑝 has a parallel extension to a neighborhood of 𝑝. The proof is an
adaptation of [LeeRM, Lemma 7.8]. Choose smooth coordinates (𝑥1 , … , 𝑥𝑁 ) on
some neighborhood 𝑈 of 𝑝, such that 𝑝 has coordinates (0, … , 0) and the image of
the coordinate map is a cube in ℝ𝑁 . We define 𝜎 on 𝑈 as follows: first parallel
transport 𝜎0 along the 𝑥1 -axis; then from each point on the 𝑥1 -axis, parallel trans-
port 𝜎 along the 𝑥2 -curve through that point; and continue by induction to obtain
a section 𝜎 defined on all of 𝑈 . It is smooth because solutions to ODEs depend

For each 𝑖 = 1, … , 𝑁 let 𝑀𝑖 ⊆ 𝑈 be the slice defined by 𝑥𝑖+1 = ⋯ = 𝑥𝑁 = 0.

smoothly on their initial conditions.

We will prove by induction on 𝑗 that

(7.10) ∇𝜕1 𝜎 = ⋯ = ∇𝜕𝑗 𝜎 = 0 on 𝑀𝑗 .
For 𝑗 = 1, this is true by construction, and for 𝑗 = 𝑁, it is the statement that 𝜎 is
parallel on 𝑈 . Assuming (7.10) is true for some 𝑗 ≥ 1, we also have ∇𝜕𝑗+1 𝜎 ≡ 0 on
𝑀𝑗+1 by construction. For 1 ≤ 𝑖 ≤ 𝑗, since ∇𝜕𝑖 𝜎 = 0 on 𝑀𝑗 , if we can show that
∇𝜕𝑗+1 (∇𝜕𝑖 𝜎) = 0 on 𝑀𝑗+1 , it will follow from uniqueness of parallel transport that
∇𝜕𝑖 𝜎 = 0 on all of 𝑀𝑗+1 . By definition of the curvature,
∇𝜕𝑗+1 (∇𝜕𝑖 𝜎) = ∇𝜕𝑖 (∇𝜕𝑗+1 𝜎) + ∇[𝜕𝑗+1 ,𝜕𝑖 ] 𝜎 + Θ(𝜕1 , 𝜕2 )𝜎.

Each of the terms on the right-hand side is zero on 𝑀𝑗+1 : the first because 𝜕𝑗+1 𝜎 ≡
0 there; the second because [𝜕𝑗+1 , 𝜕𝑖 ] = 0; and the third because ∇ is flat. This
completes the induction and shows that every 𝜎0 ∈ 𝐸𝑝 has a parallel extension to a
neighborhood of 𝑝.
Now choose a basis (𝑏1 , … , 𝑏𝑘 ) for 𝐸𝑝 , and for each 𝑖 = 1, … , 𝑘, let 𝑠𝑖 be a par-
allel extension of 𝑏𝑖 to a neighborhood of 𝑝. By continuity, the sections (𝑠1 , … , 𝑠𝑘 )
will continue to be linearly independent in some neighborhood of 𝑝, so they consti-
tute a parallel local frame. □

𝐸 → 𝑀 and a connection ∇ on 𝐸, we can define a sequence of maps on bundle-

Here is another way to look at curvature. Given a smooth complex vector bundle

valued differential forms

𝐷 𝐷
ℰ 0 (𝑀; 𝐸) ⟶ ℰ 1 (𝑀; 𝐸) ⟶ ℰ 2 (𝑀; 𝐸) → ⋯ ,
called exterior covariant differentiation, as follows.
Proposition 7.11 (The Exterior Covariant Derivative). Suppose 𝑀 is a complex
manifold, 𝐸 → 𝑀 is a smooth complex vector bundle, and ∇ is a connection on 𝐸.
Then there are operators 𝐷 ∶ ℰ 𝑞 (𝑀; 𝐸) → ℰ 𝑞+1 (𝑀; 𝐸) for each 𝑞 ≥ 0 satisfying
Curvature 201

the following properties:

(i) For 𝜎 ∈ ℰ 0 (𝑀; 𝐸) = Γ(𝐸), 𝐷𝜎 = ∇𝜎.
(ii) For 𝛼 ∈ ℰ 𝑞 (𝑀) and 𝛽 ∈ ℰ 𝑞 (𝑀; 𝐸),
′

𝐷(𝛼 ∧ 𝛽) = 𝑑𝛼 ∧ 𝛽 + (−1)𝑞 𝛼 ∧ 𝐷𝛽.

(iii) 𝐷2 𝛼 = Θ ∧ 𝛼, where Θ ∈ ℰ 2 (𝑀; End(𝐸)) is the curvature form of ∇.

Riemann operator 𝜕 𝐸 . For an 𝐸-valued 0-form (smooth section of 𝐸), we simply

Proof. The proof is closely parallel to that of Proposition 4.16 for the Cauchy–

define 𝐷𝜎 = ∇𝜎. To extend this to higher-degree forms, choose a smooth local

frame (𝑠𝑗 ) for 𝐸. For a section 𝛼 ∈ ℰ 𝑞 (𝑀; 𝐸) expressed locally as 𝛼 = 𝛼 𝑗 ⊗ 𝑠𝑗
where the 𝛼 𝑗 ’s are scalar 𝑞-forms (using the summation convention), we wish to let
𝐷𝛼 be the element of ℰ 𝑞+1 (𝑀; 𝐸) whose local expression is

(7.11) 𝐷𝛼 = 𝑑𝛼 𝑗 ⊗ 𝑠𝑗 + (−1)𝑞 𝛼 𝑗 ∧ 𝐷𝑠𝑗 .

If 𝑠𝑘̃ = 𝜏𝑘 𝑠𝑗 is another local frame, we can write 𝛼 = 𝛼

̃𝑘 ⊗ 𝑠𝑘̃ with 𝜏𝑘 𝛼
̃𝑘 = 𝛼 𝑗 , and
𝑗 𝑗

use the product rule for ∇ to compute

𝑑𝛼 𝑗 ⊗ 𝑠𝑗 + (−1)𝑞 𝛼 𝑗 ∧ 𝐷𝑠𝑗
̃𝑘 ) ⊗ ((𝜏 −1 )𝑙𝑗 𝑠𝑙̃ ) + (−1)𝑞 (𝜏𝑘 𝛼
= 𝑑 (𝜏𝑘 𝛼 ̃𝑘 ) ∧ ∇((𝜏 −1 )𝑙𝑗 𝑠𝑙̃ )
𝑗 𝑗

𝑑𝜏✭
=✭ 𝑘∧✭𝛼
̃✭𝑘 ✭
⊗✭(𝜏✭ )𝑗 𝑠𝑙̃ + 𝜏𝑘 𝑑̃
𝛼 𝑘 ⊗ (𝜏 −1 )𝑙𝑗 𝑠𝑙̃
𝑗 −1 𝑙 ✭✭ 𝑗

+ (−1) 𝜏𝑘 𝛼
̃✭ ∧ 𝑑(𝜏 −1 )𝑙𝑗 ⊗ 𝑠𝑙̃ + (−1)𝑞 𝜏𝑘 𝛼
̃𝑘 ∧ (𝜏 −1 )𝑙𝑗 ∇𝑠𝑙̃
𝑞 𝑗
𝑘 ✭✭✭ ✭✭✭✭ 𝑗

= 𝑑̃
𝛼 𝑘 ⊗ 𝑠𝑘̃ + (−1)𝑞 𝛼
̃𝑘 ∧ 𝐷𝑠𝑘̃ ,
✭✭✭
✭

where the two terms involving derivatives of 𝜏𝑘 cancel because 𝑑 ((𝜏 −1 )𝑙𝑗 𝜏𝑘 ) =
𝑗 𝑗

𝑑(𝛿𝑘𝑙 ) = 0. This proves that 𝐷 is well defined, and properties (i) and (ii) follow

To prove (iii), we note first that for 𝛼 ∈ ℰ 𝑞 (𝑀) and 𝜎 ∈ ℰ 0 (𝑀; 𝐸) = Γ(𝐸),
immediately from the definition.

property (ii) implies

𝐷2 (𝛼 ∧ 𝜎) = 𝐷(𝑑𝛼 ∧ 𝜎 + (−1)𝑞 𝛼 ∧ 𝐷𝜎 )
= (−1)𝑞+1 𝑑𝛼 ∧ 𝐷𝜎 + (−1)𝑞 𝑑𝛼 ∧ 𝐷𝜎 + 𝛼 ∧ 𝐷2 𝜎
= 𝛼 ∧ 𝐷2 𝜎.

This implies two important facts: First, it shows that the action of 𝐷2 on ℰ 𝑞 (𝑀; 𝐸)
is determined by that on ℰ 0 (𝑀; 𝐸). And second, by taking 𝛼 to be a 0-form
𝑓 ∈ 𝐶 ∞ (𝑀), it shows that 𝐷2 is linear over 𝐶 ∞ (𝑀) and is thus a smooth bun-
dle homomorphism, so it suffices to check (iii) on elements of a local frame for 𝐸.
202 7. Connections

Given a smooth local frame (𝑠𝑗 ) with connection 1-forms 𝜃𝑗𝑘 , we have

𝐷2 𝑠𝑗 = 𝐷(𝜃𝑗𝑘 ⊗ 𝑠𝑘 ) = 𝑑𝜃𝑗𝑘 ⊗ 𝑠𝑘 − 𝜃𝑗𝑘 ∧ 𝜃𝑘𝑙 ⊗ 𝑠𝑙 = (𝑑𝜃𝑗𝑙 + 𝜃𝑘𝑙 ∧ 𝜃𝑗𝑘 ) ⊗ 𝑠𝑙 = Θ𝑙𝑗 ⊗ 𝑠𝑙 ,

which is (iii) in this case. □

The First Real Chern Class

For any connection ∇ on a smooth complex vector bundle 𝐸 → 𝑀, the curvature
form Θ associated with ∇ is a global endomorphism-valued 2-form. Because the

global scalar 2-form, called the first Chern form of ∇, by

trace of an endomorphism is independent of the choice of basis, we can form a

𝑖
𝑐1 (∇) = tr Θ.
2𝜋
The reason for the coefficient will emerge shortly. (The adjective “first” reflects the
fact that there are also higher Chern forms that can be defined using higher-degree
polynomials in the curvature forms; but we will not need those. These forms, and
the cohomology classes they determine, were introduced by Shiing-Shen Chern in
1946 [Che46].)
Theorem 7.12. For any connection on a smooth complex vector bundle, the first
Chern form is closed, and its de Rham cohomology class is independent of the
choice of connection.

Proof. Let 𝐸 → 𝑀 be a smooth complex vector bundle and ∇ a connection on 𝐸.

𝑖 𝑗 𝑖
In terms of a smooth local frame, we have

𝑐1 (∇) = Θ𝑗 = 𝑑𝜃𝑗 + 𝜃𝑙 ∧ 𝜃𝑗𝑙 ).

𝑗 𝑗
2𝜋 2𝜋 (
Because wedge products of (scalar) 1-forms anticommute, we can write the second

𝜃𝑙 ∧ 𝜃𝑗𝑙 = −𝜃𝑗𝑙 ∧ 𝜃𝑙 = −𝜃𝑙 ∧ 𝜃𝑗𝑙 ,

𝑗 𝑗 𝑗
term above as

where the second equality follows from interchanging the dummy indices 𝑗 and 𝑙.
Thus the wedge product term is identically zero, so in the domain of the local frame

𝑖
we have
𝑐1 (∇) = 𝑑𝜃𝑗 ,
𝑗
2𝜋
(7.12)

because the 1-forms 𝜃𝑗𝑘 are not globally defined.)

which is locally exact and thus closed. (It is typically not globally exact, though,

suppose ∇̃ is another connection on 𝐸. Define the difference tensor between ∇ ̃

To see that its cohomology class is independent of the choice of connection,

and ∇ as the map 𝒟 ∶ Γ(𝑇ℂ 𝑀) × Γ(𝐸) → Γ(𝐸) given by

̃ 𝑋 𝜎 − ∇𝑋 𝜎.
𝒟 (𝑋)𝜎 = ∇
The First Real Chern Class 203

A straightforward computation shows that 𝒟 is linear over 𝐶 ∞ (𝑀) in both argu-

ments, so 𝒟 is actually a global endomorphism-valued 1-form, that is, a section of
ℰ 1 (End(𝐸)). It follows that its trace is a global scalar 1-form, and we have
̃ − 𝑐1 (∇) = 𝑖 ( tr Θ
𝑐1 (∇) ̃ − tr Θ) = 𝑖 ( tr 𝑑 𝜃 ̃ − tr 𝑑𝜃 ) = 𝑖 (tr 𝑑𝒟 )
2𝜋 2𝜋 2𝜋
𝑖
= 𝑑(tr 𝒟 ).
2𝜋
̃ and 𝑐1 (∇) differ by an exact form, so they define the same de Rham
Thus 𝑐1 (∇)
cohomology class. □

factor of 𝑖 is included in the definition of the Chern form.

We need one more property of this cohomology class, which explains why the

Proposition 7.13. Suppose 𝐸 → 𝑀 is a smooth complex vector bundle with a

Hermitian fiber metric. If ∇ is a metric connection on 𝐸, then 𝑐1 (∇) is a real 2-
form.

Proof. Let ∇ be a metric connection on 𝐸. In a neighborhood of each point, we may

choose an orthonormal local frame (𝑠𝑗 ), and let 𝜃𝑗𝑘 be the corresponding connection
1-forms. Proposition 7.1 shows that

𝜃𝑗𝑘 + 𝜃𝑘 = 0.
𝑗

Taking the trace of this equation, we find that the scalar 1-form 𝜃𝑗 is purely imagi-
𝑗

nary, and thus so is Θ𝑗 = 𝑑𝜃𝑗 . Thus 𝑐1 (∇) = Θ

𝑗 𝑗 𝑖 𝑗
2𝜋 𝑗
is real. □

Let 𝐸 → 𝑀 be a smooth complex vector bundle. We can always choose a

Hermitian fiber metric on ∇ and a connection ∇ compatible with it, so that 𝑐1 (∇) is
represented by a real 2-form. Thus the cohomology class determined by 𝑐1 (∇) lies
in 𝐻dR
2
(𝑀; ℝ) (considered as the subspace of 𝐻dR 2
(𝑀; ℂ) consisting of cohomol-

class of 𝑬, denoted by 𝑐1ℝ (𝐸) ∈ 𝐻dR

2
(𝑀; ℝ), to be the cohomology class of 𝑐1 (∇),
ogy classes that are invariant under conjugation). We define the first real Chern

where ∇ is any connection on 𝐸. Under the de Rham–Weil isomorphisms, we can

also consider it as an element of the sheaf cohomology group 𝐻 2 (𝑀; ℝ) or the
singular cohomology group 𝐻Sing2
(𝑀; ℝ).

Line Bundles
Now we examine how this looks in the case of line bundles. Let 𝐿 → 𝑀 be
a smooth complex line bundle, and let ∇ be a connection on 𝐿. The endomor-

is a canonical global frame, so we can consider the connection form Θ as a global

phism bundle End(𝐿) is a trivial line bundle because the identity endomorphism

scalar-valued 2-form. If we choose a trivializing cover 𝒰 = {𝑈𝛼 }𝛼∈𝐴 with a local

frame 𝑠𝛼 (that is, a nonvanishing local section) on each set 𝑈𝛼 , the connection is
204 7. Connections

determined in 𝑈𝛼 by a 1×1 matrix of 1-forms, that is, an ordinary scalar 1-form. Let
us denote the 1-form associated with the local frame 𝑠𝛼 by 𝜃𝛼 , so ∇𝑠𝛼 = 𝜃𝛼 ⊗ 𝑠𝛼 .
When 𝑈𝛼 and 𝑈𝛽 overlap, equation (3.4) shows that the local frames 𝑠𝛼 and 𝑠𝛽
are related by 𝑠𝛽 = 𝜏𝛼𝛽 𝑠𝛼 , so the transition formula (7.5) for the connection forms
becomes
(7.13) 𝜃𝛽 = 𝜏𝛼𝛽
−1
𝑑𝜏𝛼𝛽 + 𝜃𝛼
(because 1 × 1 matrices commute).
The curvature form is the globally defined 2-form Θ that satisfies Θ|𝑈𝛼 = 𝑑𝜃𝛼
for each 𝛼; and the Chern form is 𝑐1 (∇) = 2𝜋𝑖 Θ, a globally defined closed 2-form,
which is real if ∇ is a metric connection.
The next theorem explains the relationship between the first real Chern class

reason the factor of 𝑖/(2𝜋) is included in the definition of 𝑐1 (∇), and the negative
of a line bundle and the sheaf-theoretic Chern class defined in Chapter 5. It is the

sign in the definition of 𝑐(𝐿). (See the remark following the proof for a comment
about the sign.)
Theorem 7.14. Let 𝐿 → 𝑀 be a smooth complex line bundle. Under the compo-

𝐻 2 (𝑀; ℤ) → 𝐻 2 (𝑀; ℝ) → 𝐻dR (𝑀; ℝ),

sition
2

in which the first map is the homomorphism induced by the sheaf inclusion 𝜄 ∶ ℤ ↪
ℝ and the second is the inverse of the de Rham–Weil isomorphism, the sheaf-
theoretic Chern class 𝑐(𝐿) maps to the first real Chern class 𝑐1ℝ (𝐿).

theoretic Chern class map 𝑐 ∶ 𝐻 1 (𝑀; ℰ ∗ ) → 𝐻 2 (𝑀; ℤ), and second, the isomor-
Proof. This is a matter of unwinding the definitions of two maps: first, the sheaf-

phism ℛ2 ∶ 𝐻dR 2
(𝑀; ℝ) → 𝐻 2 (𝑀; ℝ) given by the de Rham–Weil theorem.
We begin by choosing an open cover 𝒰 = {𝑈𝛼 }𝛼∈𝐴 of 𝑀 by convex geodesic
balls with respect to some Riemannian metric, so that each 𝑈𝛼 and all finite inter-
sections 𝑈𝛼0 ∩⋯∩𝑈𝛼𝑝 are contractible, as in Problem 6-3. By choosing the geodesic
balls small enough, we can also ensure that over each 𝑈𝛼 there is a smooth local
frame 𝑠𝛼 (that is, a nonvanishing local section) for 𝐿.
Recall that the sheaf-theoretic Chern class is defined by 𝑐(𝐿) = −𝛿∗ ([𝐿]),
where 𝛿∗ ∶ 𝐻 1 (𝑀; ℰ ∗ ) → 𝐻 2 (𝑀; ℤ) is the connecting homomorphism in the long
exact sequence arising from the short exact sheaf sequence
𝜄 𝜀
0 → ℤ ↪ ℰ ⟶ ℰ ∗ → 0.
The homomorphism 𝛿∗ is characterized by (6.6). To compute it explicitly, we begin
by letting 𝜏𝛼𝛽 ∶ 𝑈𝛼 ∩ 𝑈𝛽 → GL(1, ℂ) be the transition functions for the given local
frames; taken together, they define a cocycle 𝜏 ∈ 𝐶 1 (𝒰; ℰ ∗ ) that represents [𝐿] ∈
𝐻 1 (𝑀; ℰ ∗ ). Because 𝑈𝛼 ∩ 𝑈𝛽 is contractible, the nonvanishing complex-valued
The First Real Chern Class 205

function 𝜏𝛼𝛽 has a complex logarithm there by the result of Problem 5-11, so we
can choose a smooth function 𝑏𝛼𝛽 ∶ 𝑈𝛼 ∩ 𝑈𝛽 → ℂ such that 𝜏𝛼𝛽 = 𝜀(𝑏𝛼𝛽 ) = 𝑒2𝜋𝑖𝑏𝛼𝛽 .
Let 𝑏 ∈ 𝐶 1 (𝒰; ℰ ) denote the cochain defined by these functions. Our construction
ensures that 𝜀# 𝑏 = 𝜏, where 𝜀# ∶ 𝐶 1 (𝒰; ℰ ) → 𝐶 1 (𝒰; ℰ ∗ ) is the homomorphism

Now consider 𝛿𝑏 ∈ 𝐶 2 (𝒰; ℰ ):

defined by (6.2).

(𝛿𝑏)𝛼𝛽𝛾 = (𝑏𝛽𝛾 − 𝑏𝛼𝛾 + 𝑏𝛼𝛽 )|𝑈 .

𝛼 ∩𝑈𝛽 ∩𝑈𝛾

The fact that 𝜏𝛼𝛽 𝜏𝛽𝛾 = 𝜏𝛼𝛾 implies that (𝛿𝑏)𝛼𝛽𝛾 is integer-valued, and because 𝑈𝛼 ∩
𝑈𝛽 ∩ 𝑈𝛾 is connected, this continuous integer-valued function is constant. Thus the
assignment

𝑘𝛼𝛽𝛾 = (𝑏𝛽𝛾 − 𝑏𝛼𝛾 + 𝑏𝛼𝛽 )|𝑈

𝛼 ∩𝑈𝛽 ∩𝑈𝛾
(7.14)

defines a 2-cocyle 𝑘 ∈ 𝐶 2 (𝒰; ℤ) that satisfies 𝜄# 𝑘 = 𝛿𝑏. We see from (6.6) that
𝛿∗ [𝜏] is represented by 𝑘, so the sheaf-theoretic Chern class 𝑐(𝐿) is represented by
−𝑘.
On the other hand, Problem 6-3 shows how the isomorphism ℛ2 ∶ 𝐻dR 2
(𝑀; ℝ)
→ 𝐻 (𝑀; ℝ) is constructed. Starting with a global closed 2-form 𝜂, on each 𝑈𝛼 we
2

need to find a 1-form 𝜑𝛼 such that 𝑑𝜑𝛼 = 𝜂|𝑈𝛼 , and on each nonempty intersection
𝑈𝛼 ∩ 𝑈𝛽 a smooth function 𝑢𝛼𝛽 such that 𝜑𝛽 |𝑈𝛼 ∩𝑈𝛽 − 𝜑𝛼 |𝑈𝛼 ∩𝑈𝛽 = 𝑑𝑢𝛼𝛽 ; and then
ℛ2 ([𝜂]) = [[𝑎]] where 𝑎𝛼𝛽𝛾 is the restriction to 𝑈𝛼 ∩𝑈𝛽 ∩𝑈𝛾 of the constant function
𝑢𝛽𝛾 − 𝑢𝛼𝛾 + 𝑢𝛼𝛽 .
Let ∇ be a connection on 𝐿 that is compatible with some Hermitian fiber metric,
and for each 𝛼 let 𝜃𝛼 be the connection 1-form on 𝑈𝛼 ; the proof of Proposition 7.13
shows that 𝜃𝛼 is purely imaginary. Let Θ be the connection form of ∇, and let
𝜂 = 2𝜋𝑖 Θ be its first Chern form. On each set 𝑈𝛼 , we have Θ|𝑈𝛼 = 𝑑𝜃𝛼 , so to apply
the de Rham isomorphism to 𝜂 we can take 𝜑𝛼 to be the real 1-form 𝑖
𝜃
2𝜋 𝛼
on 𝑈𝛼 .
On each nonempty overlap 𝑈𝛼 ∩ 𝑈𝛽 , (7.13) shows that

𝜃𝛽 − 𝜃𝛼 = 𝜏𝛼𝛽
−1
𝑑𝜏𝛼𝛽 .

Using the same functions 𝑏𝛼𝛽 as above, we have 𝜏𝛼𝛽 = 𝑒2𝜋𝑖𝑏𝛼𝛽 , so

𝑖
𝜑𝛽 − 𝜑 𝛼 = (𝜃 − 𝜃𝛼 ) = −𝑑𝑏𝛼𝛽 ,
2𝜋 𝛽
and we can take 𝑢𝛼𝛽 = −𝑏𝛼𝛽 . This yields ℛ2 [𝑐1 (∇)] = [[𝑎]], where 𝑎 is the ℂ-
valued 2-cocycle given by

𝑎𝛼𝛽𝛾 = −(𝑏𝛽𝛾 − 𝑏𝛼𝛾 + 𝑏𝛼𝛽 )|𝑈 .

𝛼 ∩𝑈𝛽 ∩𝑈𝛾
(7.15)
206 7. Connections

Comparing (7.14) and (7.15), we see that 𝑎 = −𝜄# 𝑘, so 𝜄∗ 𝑐(𝐿) = −[[𝜄# 𝑘]] = [[𝑎]] =
ℛ2 ([𝑐1 (∇)]). □

theoretic Chern class as 𝑐(𝐿) = 𝛿 ∗ ([𝐿]), without the negative sign. In some cases,
Remark. As we noted in Chapter 6, some authors write the definition of the sheaf-

tions, with (3.4) replaced by 𝑠𝛼 = 𝜏𝛼𝛽 𝑠𝛽 . But in other cases, it seems to be based
this may just be a matter of choosing a different convention for the transition func-

from an incorrect transformation equation between the connection forms 𝜃𝛽 and 𝜃𝛼 :

on a computational error. For example, in [GH94, p. 141], the discrepancy results

their transition functions 𝑔𝛼𝛽 play the same role as our 𝜏𝛼𝛽 , but their transformation
formula reverses the roles of 𝜃𝛼 and 𝜃𝛽 compared to our equation (7.13), resulting
in an incorrect sign for the image of 𝑐1 (∇) in 𝐻 2 (𝑀; ℤ). Be sure to check all such
computations carefully before using them.

Corollary 7.15. Suppose 𝑀 is a connected compact Riemann surface, 𝐿 → 𝑀 is

a smooth complex line bundle, and Ω is a closed 2-form representing 𝑐1ℝ (𝐿). The
degree of 𝐿 is given by the formula

deg(𝐿) = Ω.
∫𝑀

Proof. Let 𝜑 ∈ Sing2 (𝑀; ℤ) be a singular cocycle representing the sheaf-theoretic

Chern class 𝑐(𝐿) ∈ 𝐻Sing
2
(𝑀; ℤ), and let 𝜇 be a singular cycle representing the
fundamental class [𝑀] ∈ 𝐻2 (𝑀). Then by definition deg(𝐿) = 𝜑(𝜇). Because
𝑀 is a smooth oriented compact manifold, we can choose 𝜇 to be a smooth chain
obtained from a smooth triangulation of 𝑀.
Now the image of 𝑐(𝐿) in 𝐻Sing
2
(𝑀; ℂ) under the coefficient homomorphism
ℤ ↪ ℂ is represented by the same cocycle 𝜑, but now thought of as a homo-
morphism from Sing2 (𝑀) into ℂ. Similarly, its image in 𝐻Sing,∞
2
(𝑀; ℂ) is again
the same cocycle, but with its action restricted to smooth chains. On the other

𝜑
̃ ∈ Sing2,∞ (𝑀; ℂ) given by 𝜑̃ (𝑐) = ∫𝑐 Ω. Since integrating over 𝜇 is the same as
hand, Theorem 7.14 shows that this cocycle is also represented by the cocycle

integrating over 𝑀 [LeeSM, Prop. 16.8], the result follows. □

It is important to realize that the coefficient homomorphism 𝐻 2 (𝑀; ℤ) →

𝐻 (𝑀; ℂ) (or equivalently 𝐻Sing
2 2
(𝑀; ℤ) → 𝐻Sing2
(𝑀; ℂ)) need not be injective,

Chern class. For example, if 𝑀 is any smooth manifold for which 𝐻Sing 2
(𝑀; ℤ)
so the sheaf-theoretic Chern class may contain more information than the first real

contains a nontrivial torsion element 𝛾 (meaning 𝑘𝛾 = 0 for some positive integer

𝑘), then the image of 𝛾 in 𝐻dR 2
(𝑀; ℝ) is zero. (Simple examples of such manifolds
are the real projective spaces ℝℙ𝑛 for 𝑛 ≥ 2.) By Theorem 6.29, there is a nontriv-
ial smooth complex line bundle over 𝑀 whose sheaf-theoretic Chern class is equal
to 𝛾; but its first real Chern class is zero.
The Chern Connection 207

The Chern Connection

Now we turn our attention to holomorphic vector bundles on complex manifolds.
We can always choose a Hermitian fiber metric on such a bundle, and as we men-
tioned earlier, there is a connection on the bundle that is compatible with the metric.
As you can see from the use of a partition of unity in the construction of such con-
nections, such a connection is far from unique. But when the bundle is holomorphic,
we can require an additional condition that will guarantee uniqueness.
It is useful to consider the case of Riemannian metrics for comparison. On
a Riemannian manifold, there are many connections on the tangent bundle that
are compatible with the Riemannian metric; but there is a unique connection, the

meaning that ∇𝑋 𝑌 − ∇𝑌 𝑋 = [𝑋, 𝑌 ] for all smooth vector fields 𝑋 and 𝑌 . But the
Levi-Civita connection, that satisfies the additional condition of being torsion-free,

torsion-free condition makes sense only for connections on the tangent bundle, so
we need another condition to determine a unique connection on a holomorphic bun-

Suppose 𝑀 is a complex manifold and 𝐸 → 𝑀 is a holomorphic vector bundle.

dle.

Using the decomposition 𝑇ℂ∗ 𝑀 = Λ1ℂ 𝑀 = Λ1,0 𝑀 ⊕ Λ0,1 𝑀, we can decompose

a connection on 𝐸 as ∇ = ∇(1,0) + ∇(0,1) , where ∇(1,0) 𝜎 ∈ Γ(Λ1,0 𝑀 ⊗ 𝐸) and
∇(0,1) 𝜎 ∈ Γ(Λ0,1 𝑀 ⊗ 𝐸).

A connection on a holomorphic vector bundle 𝐸 → 𝑀 is said to be compatible

Here is the additional condition we wish to impose in the holomorphic case.

with the holomorphic structure if ∇(0,1) is equal to the operator 𝜕 𝐸 defined by

Proposition 4.16.
Proposition 7.16 (Compatibility with the Holomorphic Structure). Let 𝐸 →
𝑀 be a holomorphic vector bundle and ∇ a connection on 𝐸. The following are
equivalent:
(a) ∇ is compatible with the holomorphic structure (i.e., ∇(0,1) = 𝜕 𝐸 ).
(b) Whenever 𝜎 is a holomorphic local section of 𝐸 and 𝑍 is a smooth local
section of 𝑇 ″ 𝑀, we have ∇𝑍 𝜎 = 0.
(c) For each holomorphic local frame (𝑠𝑗 ), we have ∇𝑠𝑗 = 𝜃𝑗𝑘 ⊗ 𝑠𝑘 where the
1-forms 𝜃𝑗𝑘 are all of type (1, 0).

holomorphic local frame (𝑠𝑗 ). Let 𝜃𝑗𝑘 be the connection 1-forms with respect to this
Proof. This is a local issue, so we may work in an open set over which there is a

frame. Taking the projection of both sides of (7.4) onto Λ0,1 𝑀 ⊗ 𝐸, we have
∇(0,1) (𝜎 𝑗 𝑠𝑗 ) = 𝜕𝜎 𝑗 ⊗ 𝑠𝑗 + 𝜎 𝑗 (𝜃𝑗𝑘 )(0,1) ⊗ 𝑠𝑘 .
On the other hand, (4.20) shows that
𝜕 𝐸 (𝜎 𝑗 𝑠𝑗 ) = 𝜕𝜎 𝑗 ⊗ 𝑠𝑗 .
208 7. Connections

Comparing these two equations, we see that ∇(0,1) = 𝜕 𝐸 if and only if (𝜃𝑗𝑘 )(0,1) = 0
for all 𝑗 and 𝑘, which shows that (a) ⇔ (c).
To prove that (c) ⇒ (b), suppose (c) holds. Let 𝜎 be a holomorphic local section
of 𝐸 and 𝑍 be a smooth local section of 𝑇 ″ 𝑀. We can write 𝜎 = 𝜎 𝑗 𝑠𝑗 , where now
the component functions 𝜎 𝑗 are holomorphic, and compute
∇𝑍 𝜎 = 𝑍(𝜎 𝑗 )𝑠𝑗 + 𝜃𝑗𝑘 (𝑍)𝑠𝑘 = 0 + 0.
Conversely if (b) holds, then for any local section 𝑍 of 𝑇 ″ 𝑀,
0 = ∇𝑍 𝑠𝑗 = 𝜃𝑗𝑘 (𝑍)𝑠𝑘 .

This shows that each form 𝜃𝑗𝑘 vanishes on 𝑇 ″ 𝑀, which is equivalent to being of
type (1, 0), so (b) ⇒ (c). □

Here is the fundamental fact about connections on Hermitian holomorphic bun-

dles.
Theorem 7.17 (Chern Connection Theorem). On every Hermitian holomorphic
vector bundle there is a unique connection, called the Chern connection, that is
compatible with the metric and the holomorphic structure.

Proof. We begin by proving uniqueness. Suppose ∇ is such a connection on 𝐸 →

𝑀, and let (𝑠𝑗 ) be a holomorphic local frame for 𝐸 over an open subset 𝑈 ⊆ 𝑀.
Writing ∇𝑠𝑗 = 𝜃𝑗𝑘 ⊗ 𝑠𝑘 , we note first that Proposition 7.16 implies that the forms
𝜃𝑗𝑘 are all of type (1, 0). If we let ℎ𝑗𝑘 = ⟨𝑠𝑗 , 𝑠𝑘 ⟩, then compatibility with the metric
implies the following equation for every local section 𝑍 of 𝑇 ′ 𝑀:
𝑍(ℎ𝑗𝑘 ) = 𝑍⟨𝑠𝑗 , 𝑠𝑘 ⟩ = ⟨∇𝑍 𝑠𝑗 , 𝑠𝑘 ⟩ + ⟨𝑠𝑗 , ∇𝑍 𝑠𝑘 ⟩
(7.16) = ⟨𝜃𝑗𝑙 (𝑍)𝑠𝑙 , 𝑠𝑘 ⟩ + 0
= 𝜃𝑗𝑙 (𝑍)ℎ𝑙𝑘 .

Since the matrix (ℎ𝑗𝑘 ) is positive definite, it has an inverse matrix, denoted by (ℎ𝑗𝑘 ).
Multiplying both sides of (7.16) by ℎ𝑘𝑚 and simplifying, we obtain
ℎ𝑘𝑚 𝑍(ℎ𝑗𝑘 ) = 𝜃𝑗𝑚 (𝑍).
Since this is true for every section 𝑍 of 𝑇 ′ 𝑀, it implies
(7.17) 𝜃𝑗𝑚 = ℎ𝑘𝑚 𝜕ℎ𝑗𝑘 .
This shows that ∇ is uniquely determined if it exists.
To prove existence, we use (7.17) to define ∇ in terms of each holomorphic lo-
cal frame. These forms are of type (1, 0) by definition, so the resulting connection

putation above (together with the analogous computation for 𝑍) shows that it is
is compatible with the holomorphic structure; and reverse-engineering the com-
The Chern Connection 209

associated with different local frames agree where they overlap, so ∇ is globally
also compatible with the metric. Then uniqueness guarantees that the definitions

defined. □

The relative simplicity of (7.17) should be compared with the much more com-
plicated formula for the connection coefficients of the Levi–Civita connection on a
Riemannian manifold [LeeRM, eq. (5.12)].
Proposition 7.18 (Curvature of the Chern Connection is Type (1, 1)). Suppose
𝑀 is a complex manifold and 𝐸 → 𝑀 is a Hermitian holomorphic vector bun-
dle. The curvature form Θ associated with the Chern connection on 𝐸 lies in
ℰ 1,1 (𝑀; End(𝐸)).

Proof. In terms of a holomorphic local frame (𝑠𝑗 ), we can write

(7.18) 𝑗 = 𝜕𝜃𝑗 + 𝜕𝜃𝑗 + 𝜃𝑘 ∧ 𝜃𝑗 .
Θ𝑚 𝑚 𝑚 𝑚 𝑘

Because the forms 𝜃𝑗𝑚 are all of type (1, 0), the first term on the right-hand side of
(7.18) is of type (1, 1) and the last two are of type (2, 0); thus it suffices to show that
the last two terms sum to zero. This is just a computation using (7.17):
𝜕𝜃𝑗𝑚 + 𝜃𝑘𝑚 ∧ 𝜃𝑗𝑘 = 𝜕 (ℎ𝑘𝑚 𝜕ℎ𝑗𝑘 ) + (ℎ𝑝𝑚 𝜕ℎ𝑘𝑝 ) ∧ (ℎ𝑞𝑘 𝜕ℎ𝑗𝑞 )
= 𝜕ℎ𝑘𝑚 ∧ 𝜕ℎ𝑗𝑘 + ℎ𝑝𝑚 ℎ𝑞𝑘 𝜕ℎ𝑘𝑝 ∧ 𝜕ℎ𝑗𝑞 .
Now differentiating ℎ𝑘𝑝 ℎ𝑝𝑙 = 𝛿𝑙𝑘 and multiplying by ℎ𝑙𝑚 yields
𝜕ℎ𝑘𝑚 = −ℎ𝑙𝑚 ℎ𝑘𝑝 𝜕ℎ𝑝𝑙 ,
and substituting this above and renaming the dummy indices proves the result. □

The Chern Connection on a Line Bundle

and its curvature simplify even further. Let 𝐿 → 𝑀 be a Hermitian holomorphic

In the case of a holomorphic line bundle, the formulas for the Chern connection

line bundle and let ∇ be its Chern connection. Given a local holomorphic frame 𝑠 for
𝐿 (that is, a nonvanishing holomorphic local section) over an open set 𝑈 ⊆ 𝑀, let 𝜃
be the corresponding connection 1-form. The fiber metric is completely determined
in 𝑈 by the strictly positive smooth function ℎ = |𝑠|2 = ⟨𝑠, 𝑠⟩. In this situation,
(7.17) reduces to
(7.19) 𝜃 = ℎ−1 𝜕ℎ = 𝜕(log ℎ),
and its curvature is the globally defined 2-form Θ whose expression in terms of each
holomorphic local frame 𝑠 is
(7.20) Θ = 𝑑𝜃 = 𝜕𝜕(log ℎ) = 𝜕𝜕 ( log |𝑠|2 ).

𝑖 𝑖
Thus the Chern form for this connection has the local expression
𝑐1 (∇) = 𝜕𝜕(log ℎ) = 𝜕𝜕(log |𝑠|2 ).
2𝜋 2𝜋
210 7. Connections

Next we will show how these formulas allow us to easily compute the connec-
tion and curvature forms for various related line bundles—tensor product bundles,

product of line bundles 𝐿 ⊗ 𝐿′ , having chosen Hermitian fiber metrics for 𝐿 and
dual bundles, and pullback bundles. We begin with tensor products. For a tensor

𝐿′ , we obtain a fiber metric for 𝐿 ⊗ 𝐿′ , called the tensor product metric, by setting
|𝜎 ⊗ 𝜎 ′ |2 = |𝜎|2 |𝜎 ′ |2 .
Proposition 7.19 (Curvature of a Tensor Product of Line Bundles). Let 𝐿, 𝐿′ →
𝑀 be Hermitian holomorphic line bundles, and let Θ𝐿 and Θ𝐿′ be curvature forms

of the Chern connection on 𝐿 ⊗ 𝐿′ is given by

of their Chern connections. With respect to the tensor product metric, the curvature

Θ𝐿⊗𝐿′ = Θ𝐿 + Θ𝐿′ .

Proof. With respect to holomorphic local frames 𝑠 for 𝐿 and 𝑠′ for 𝐿′ , if we set
ℎ = |𝑠|2 and ℎ′ = |𝑠′ |2 , then the fiber metric on 𝐿 ⊗ 𝐿′ is represented locally by
the function ℎℎ′ = |𝑠 ⊗ 𝑠′ |2 . Thus the curvature form of the Chern connection on
𝐿 ⊗ 𝐿′ is given by
Θ𝐿⊗𝐿′ = 𝜕𝜕(log ℎℎ′ ) = 𝜕𝜕(log ℎ + log ℎ′ ) = Θ𝐿 + Θ𝐿′ . □

Next we consider the dual bundle.

Proposition 7.20 (Curvature of the Dual Bundle). Let 𝐿 → 𝑀 be a Hermitian
holomorphic line bundle and let Θ𝐿 be the curvature form of its Chern connection.
When 𝐿∗ is endowed with the dual metric (see Problem 7-3), the curvature of its

Θ𝐿∗ = −Θ𝐿 .
Chern connection is given by

Proof. This follows easily from the results of Problems 7-4 and 7-5. □

Finally, we consider pullbacks of line bundles. Suppose 𝑀 and 𝑀 ′ are complex

manifolds and 𝐿 → 𝑀 is a holomorphic line bundle. Given a holomorphic map
𝑓 ∶ 𝑀 ′ → 𝑀, recall that the pullback bundle 𝑓 ∗ 𝐿 → 𝑀 ′ is defined as
𝑓 ∗ 𝐿 = {(𝑥, 𝑣) ∈ 𝑀 ′ × 𝐿 ∶ 𝑣 ∈ 𝐿𝑓 (𝑥) }.
Thus the fiber of 𝑓 ∗ 𝐿 over 𝑥 ∈ 𝑀 ′ is {𝑥} × 𝐿𝑓 (𝑥) , which we can canonically
identify with 𝐿𝑓 (𝑥) . Using this identification, we can define a fiber metric on 𝑓 ∗ 𝐿,
called the pullback metric, by
̃ 𝑓 ∗ 𝐿 = ⟨𝑣, 𝑤⟩𝐿
⟨𝑣,̃ 𝑤⟩ ̃ = (𝑥, 𝑤) ∈ (𝑓 ∗ 𝐿)𝑥 .
for 𝑣 ̃ = (𝑥, 𝑣), 𝑤
For any smooth local frame 𝑠 for 𝐿, we get a smooth local frame 𝑓 ∗ 𝑠 for 𝑓 ∗ 𝐿
defined by 𝑓 ∗ 𝑠(𝑥) = (𝑥, 𝑠(𝑓 (𝑥))), and the pullback metric satisfies |𝑓 ∗ 𝑠(𝑥)|𝑓 ∗ 𝐿 =
|𝑠(𝑓 (𝑥))|𝐿 , so it is smooth by composition.
The Chern Connection 211

Proposition 7.21 (Curvature of a Pullback Bundle). Let 𝑀 and 𝑀 ′ be complex

manifolds, 𝐿 → 𝑀 a holomorphic line bundle, and 𝑓 ∶ 𝑀 ′ → 𝑀 a holomorphic
map. With respect to any Hermitian fiber metric on 𝐿 and the pullback metric on
𝑓 ∗ 𝐿, the curvatures of the Chern connections on 𝐿 and 𝑓 ∗ 𝐿 satisfy
Θ𝑓 ∗ 𝐿 = 𝑓 ∗ Θ𝐿 .

Proof. Given 𝑝 ∈ 𝑀 ′ , let 𝑠 be a holomorphic local frame for 𝐿 in a neighborhood

of 𝑓 (𝑝), and let 𝑠′ = 𝑓 ∗ 𝑠, which is a holomorphic local frame for 𝑓 ∗ 𝐿. If we write
ℎ = |𝑠|2 and ℎ′ = |𝑠′ |2 , then ℎ′ = ℎ ∘ 𝑓 , and therefore the curvature form of the
Chern connection on 𝑓 ∗ 𝐿 is given locally by
Θ𝑓 ∗ 𝐿 = 𝜕𝜕(log ℎ′ ) = 𝜕𝜕(log 𝑓 ∗ ℎ) = 𝑓 ∗ 𝜕𝜕(log ℎ) = 𝑓 ∗ Θ𝐿 .
Since this is true in a neighborhood of each point, it follows that Θ𝑓 ∗ 𝐿 = 𝑓 ∗ Θ𝐿
globally. □

bundles on Riemann surfaces. Recall that if 𝐿 → 𝑀 is a holomorphic line bundle

The next theorem gives an extremely useful way to compute the degrees of line

on a compact Riemann surface and 𝜎 is a meromorphic section of 𝐿, the divisor

of 𝝈, denoted by (𝜎), is the formal sum of zeros and poles of 𝜎 with coefficients
indicating their orders (positive for zeros, negative for poles). If 𝐷 = ∑𝑘𝛼=1 𝑚𝛼 𝑝𝛼
is a divisor, we define the degree of 𝑫 to be the sum of its coefficients, that is,
𝑘
deg 𝐷 = 𝑚𝛼 .
∑
𝛼=1

Theorem 7.22 (Degree of a Line Bundle Associated with a Divisor). Let 𝑀 be

a connected compact Riemann surface. For any divisor 𝐷 on 𝑀, the degree of the
line bundle 𝐿𝐷 is equal to the degree of the divisor 𝐷.

Proof. Let 𝐷 = ∑𝑘𝛼=1 𝑚𝛼 𝑝𝛼 be a divisor on 𝑀. Proposition 3.41 shows that 𝐿𝐷

has a meromorphic section 𝜎 whose divisor is 𝐷. Choose any Hermitian metric on
𝐿, and let ∇ be the corresponding Chern connection and Θ ∈ ℰ 1,1 (𝑀) its curvature

For each point 𝑝𝛼 , choose a holomorphic coordinate 𝑧𝛼 on an open disk 𝑈𝛼

form.

centered at 𝑝𝛼 . For 𝜀 > 0, let 𝑈𝛼 (𝜀) ⊆ 𝑀 be the set {𝑞 ∈ 𝑈𝛼 ∶ |𝑧𝛼 (𝑞)| < 𝜀}. For
any 𝜀 small enough that 𝑈 𝛼 (𝜀) ⊆ 𝑈𝛼 for each 𝛼 and the sets 𝑈 1 (𝜀), … , 𝑈 𝑘 (𝜀) are
disjoint, let 𝑀𝜀 = 𝑀 ∖ (𝑈1 (𝜀) ∪ ⋯ ∪ 𝑈𝑘 (𝜀)).
On 𝑀 ∖ {𝑝1 , … , 𝑝𝑘 }, the section 𝜎 is holomorphic and nonvanishing, so it can
be used as a holomorphic frame for 𝐿 with which to compute Θ on that set. Formula
(7.20) shows that on that set we have
Θ = 𝜕𝜕 ( log |𝜎|2 ) = 𝑑𝜕 ( log |𝜎|2 ).
212 7. Connections

Thus by Stokes’s theorem,

𝑘
Θ= 𝜕 ( log |𝜎|2 ) = − 𝜕 ( log |𝜎|2 ),
∫𝑀𝜀 ∫𝜕𝑀𝜀
∑∫𝜕𝑈
𝛼=1 𝛼 (𝜀)

orientation on 𝜕𝑈 𝛼 (𝜀), considered as the boundary of 𝑈 𝛼 (𝜀), is opposite to its ori-

where the minus sign in the last expression comes from the fact that the Stokes

entation considered as part of 𝜕𝑀𝜀 . It follows from Corollary 7.15 that

𝑘
𝑖 𝑖
deg(𝐿) = Θ = − lim 𝜕 ( log |𝜎|2 ).
2𝜋 ∫ 𝜀→0 ∑ 2𝜋 ∫
(7.21)
𝑀 𝛼=1 𝜕𝑈 𝛼 (𝜀)

Now consider one of the disks 𝑈 𝛼 (𝜀), and write the corresponding holomorphic
coordinate as 𝑧 = 𝑧𝛼 for convenience. Choose a nonvanishing holomorphic section
𝑠 of 𝐿 on 𝑈𝛼 , and write 𝜎(𝑧) = 𝑧𝑚𝛼 𝑓 (𝑧)𝑠(𝑧), where 𝑚𝛼 is the coefficient of 𝑝𝛼 in
𝐷 and 𝑓 is a nonvanishing holomorphic function on 𝑈 𝛼 (𝜀). Let ℎ(𝑧) = |𝑠(𝑧)|2 , a
nonvanishing smooth function. In these coordinates, since |𝜎|2 does not vanish on
𝜕𝑈 𝛼 (𝜀), we have
𝜕 (𝑧𝑚𝛼 𝑧𝑚𝛼 |𝑓 (𝑧)|2 ℎ(𝑧)))
𝜕 ( log |𝜎|2 ) =
𝑧𝑚𝛼 𝑧𝑚𝛼 |𝑓 (𝑧)|2 ℎ(𝑧)
𝑑𝑧
= 𝑚𝛼 + 𝑢(𝑧) 𝑑𝑧,
𝑧
where 𝑢 is a smooth function of 𝑧 on 𝑈 𝛼 (𝜀).
We can parametrize 𝜕𝑈 𝛼 (𝜀) in 𝑧-coordinates by 𝑧 = 𝜀𝑒𝑖𝜃 for 𝜃 ∈ [0, 2𝜋], so
𝑑𝑧 = 𝑖𝜀𝑒𝑖𝜃 𝑑𝜃. This yields

𝑖 𝑖 𝑖
2𝜋 2𝜋
𝜕 ( log |𝜎|2 ) = 𝑚𝛼 𝑖𝑑𝜃 + 𝑢(𝜀𝑒𝑖𝜃 )𝑖𝜀𝑒𝑖𝜃 𝑑𝜃.
2𝜋 ∫𝜕𝑈 𝛼 (𝜀) 2𝜋 ∫0 2𝜋 ∫0

The first term on the right is equal to −𝑚𝛼 , and the second approaches zero as 𝜀 → 0.
Applying this to each term in the sum (7.21) proves the result. □

Corollary 7.23. Let 𝑀 be a connected compact Riemann surface. If 𝐿 → 𝑀 is a

holomorphic line bundle of negative degree, then 𝐿 admits no nontrivial holomor-
phic sections.

Proof. Theorem 7.22 implies the contrapositive: if 𝐿 admits a nontrivial holomor-

phic section 𝜎, then it is isomorphic to the bundle 𝐿𝐷 for 𝐷 = (𝜎); and since 𝐷 is
an effective divisor (i.e., it has only nonnegative coefficients), its degree is either
zero or a sum of positive integers. □

Corollary 7.24. Let 𝑀 be a connected compact Riemann surface and 𝑓 be a mero-

morphic function on 𝑀. Then deg(𝑓 ) = 0.
The Chern Connection 213

Proof. We can consider 𝑓 as a meromorphic section of the trivial holomorphic line

bundle, so the line bundle 𝐿𝐷 associated with the divisor 𝐷 = (𝑓 ) is trivial. Since
the trivial line bundle has degree 0, so does (𝑓 ). □

Proposition 7.25 (Characterizations of ℂℙ1 ). Let 𝑀 be a connected compact

Riemann surface. The following are equivalent.
(a) There exists a meromorphic function on 𝑀 with only a single pole of order
1.
(b) There are distinct points 𝑝, 𝑞 ∈ 𝑀 such that the point bundles 𝐿{𝑝} and
𝐿{𝑞} are isomorphic.
(c) There is a holomorphic line bundle 𝐿 → 𝑀 of degree 1 whose space of
global holomorphic sections has dimension 2.
(d) 𝑀 is biholomorphic to ℂℙ1 .

Proof. We will prove (a) ⇒ (b) ⇒ (c) ⇒ (d) ⇒ (a). First, assuming (a), let 𝑓 be a
meromorphic function with a simple pole at 𝑝 ∈ 𝑀. Corollary 7.24 implies that 𝑓
must have exactly one simple zero, say at 𝑞. Let 𝐿 denote the point bundle 𝐿{𝑝} . It
has a holomorphic section 𝜎𝑝 with a simple zero at 𝑝 by Theorem 3.39. Then 𝑓 𝜎𝑝
will have a removable singularity at 𝑝, so it is a holomorphic section of 𝐿 with a
simple zero at 𝑞. Thus Theorem 3.39 shows that 𝐿{𝑝} ≅ 𝐿{𝑞} .
Next, assuming (b), let us denote the bundle 𝐿{𝑝} ≅ 𝐿{𝑞} by 𝐿. By Theorem
3.39, there are holomorphic sections 𝑠𝑝 , 𝑠𝑞 ∈ 𝒪(𝑀; 𝐿) that vanish simply at 𝑝 and
𝑞, respectively, and nowhere else. Because they vanish at different points, they
cannot be linearly dependent. If 𝜎 ∈ 𝒪(𝑀; 𝐿) is an arbitrary nontrivial holomor-
phic section, because 𝐿 has degree 1 the divisor of 𝜎 must consist of a single point
𝑥. If 𝑥 is equal to 𝑝, then Theorem 3.41 shows that 𝜎 is a constant multiple of 𝑠𝑝 .
Otherwise, 𝑠𝑝 (𝑥) ≠ 0, so there is a complex number 𝑎 such that 𝑠𝑞 (𝑥) = 𝑎𝑠𝑝 (𝑥).
The section 𝜎 ′ = 𝑎𝑠𝑝 − 𝑠𝑞 also vanishes at 𝑥, and is not identically zero because
𝑠𝑝 and 𝑠𝑞 are linearly independent. Because deg 𝐿 = 1, Theorem 7.22 shows that
𝜎 ′ cannot vanish anywhere else in 𝑀, so it follows from Theorem 3.41 that 𝜎 is a
constant multiple of 𝜎 ′ . In either case, we have shown that 𝑠𝑝 and 𝑠𝑞 span 𝒪(𝑀; 𝐿).

Now assume 𝐿 → 𝑀 is a holomorphic line bundle of degree 1 such that

This proves (c).

dim 𝒪(𝑀; 𝐿) = 2. We will show that the associated map 𝐹 ∶ 𝑀 → ℂℙ1 is a

biholomorphism. By Lemma 3.42, 𝐹 will be a holomorphic embedding provided
𝒪(𝑀; 𝐿) separates points and directions.
Let (𝑠1 , 𝑠2 ) be a basis for 𝒪(𝑀; 𝐿). To see that they separate points, let 𝑝, 𝑞 ∈
𝑀 be arbitrary distinct points. If 𝑠1 (𝑝) = 0, then 𝑠1 (𝑞) ≠ 0 because each section
has only one simple zero. Otherwise, there is a complex number 𝑎 such that 𝑠2 (𝑝) =
𝑎𝑠1 (𝑝), and the section 𝜎 = 𝑎𝑠1 − 𝑠2 vanishes at 𝑝. As before, it can have no other
zeros, so 𝜎(𝑞) ≠ 0.
214 7. Connections

To see that they separate directions, again let 𝑝 ∈ 𝑀 be arbitrary and let 𝜎
be the section defined above. Theorem 7.22 shows that 𝜎 cannot have a zero of
order greater than 1 at 𝑝, so in terms of a local holomorphic frame 𝑠 for 𝐿 we have
𝜎 = 𝑓 𝑠 where 𝑓 (𝑝) = 0 and 𝑑𝑓𝑝 ≠ 0. Thus given any nonzero 𝑣 ∈ 𝑇𝑝′ 𝑀, we have
𝑣𝑓 (𝑝) = 𝑑𝑓𝑝 (𝑣) ≠ 0.
Therefore 𝐹 is a holomorphic embedding. Since 𝑀 and ℂℙ1 are connected
compact manifolds of the same dimension, 𝐹 is both an open and closed map and

Finally, to prove (d) ⇒ (a), just note that the function 𝑓 ∶ ℂℙ1 ∖ {[0, 1]} → ℂ
therefore surjective, so it is a biholomorphism.

defined by 𝑓 ([𝑤0 , 𝑤1 ]) = 𝑤1 /𝑤0 has a simple pole at [0, 1] and is holomorphic

elsewhere. □

Here are some examples that illustrate how to use these formulas.

Example 7.26 (Curvatures of Projective Line Bundles). Let 𝑇 → ℂℙ𝑛 be the

tautological bundle. Since each fiber is canonically identified with a line in ℂ𝑛+1 , it
inherits a Hermitian metric ⟨⋅, ⋅⟩ from the standard metric on ℂ𝑛+1 . On each open
subset 𝑈𝛼 = {[𝑤] ∈ ℂℙ𝑛 ∶ 𝑤𝛼 ≠ 0}, we have a nonvanishing section 𝑠𝛼 of 𝑇
given by (3.6), and its norm with respect to this metric satisfies

|𝑤|2
|𝑠𝛼 ([𝑤])|2 = .
|𝑤𝛼 |2

On 𝑈𝛼 , the Chern connection for this metric is given by the connection form

|𝑤|2
𝜃𝛼 = 𝜕 log |𝑠𝛼 |2 = 𝜕 log ,
|𝑤𝛼 |2

and its curvature form is

|𝑤|2
Θ𝑇 |𝑈 = 𝜕𝜕 log |𝑠𝛼 |2 = 𝜕𝜕 log .
𝛼 |𝑤𝛼 |2

For the hyperplane bundle 𝐻 = 𝑇 ∗ with the dual fiber metric, Proposition 7.20
shows that
|𝑤𝛼 |2
Θ𝐻 |𝑈 = −Θ𝑇 |𝑈 = 𝜕𝜕 log ,
𝛼 𝛼 |𝑤|2
and more generally for a tensor power 𝐻 𝑚 with 𝑚 ∈ ℤ,

|𝑤𝛼 |2
Θ𝐻 𝑚 |𝑈 = 𝑚𝜕𝜕 log .
𝛼 |𝑤|2
The Chern Connection 215

Let us compute the curvature form Θ𝐻 explicitly in terms of affine coordinates

(𝑧1 , … , 𝑧𝑛 ) ↔ [𝑧1 , … , 1, … , 𝑧𝑛 ] on 𝑈𝛼 :
1
Θ𝐻 |𝑈 = 𝜕𝜕 log = 𝜕𝜕 log(1 + |𝑧|2 )
𝛼 1 + |𝑧| 2

∑𝑗 𝑑𝑧 ∧ 𝑑𝑧𝑗 ∑𝑗,𝑘 𝑧𝑗 𝑧𝑘 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑘

𝑗
= − .
(7.22)

1 + |𝑧|2 (1 + |𝑧|2 )2
It is notable that the formula is exactly the same in all the affine charts. //
Example 7.27 (Degrees of Line Bundles on ℂℙ1 ). The holomorphic sections of
the hyperplane bundle 𝐻 → ℂℙ1 correspond to complex-linear functionals on ℂ2 ,

is a single point. Thus it follows from Theorem 7.22 that deg(𝐻) = 1. Since the
and each such section vanishes simply on a projective hyperplane, which in this case

degree map is a homomorphism from the Picard group to ℤ, for any tensor power
𝐻 𝑑 we have deg(𝐻 𝑑 ) = 𝑑.
We can also verify this using the Chern form. On ℂℙ1 , formula (7.22) reduces

(1 + |𝑧|2 )𝑑𝑧 ∧ 𝑑𝑧 − |𝑧|2 𝑑𝑧 ∧ 𝑑𝑧 𝑑𝑧 ∧ 𝑑𝑧

to
Θ 𝐻 |𝑈 = = .
𝛼 (1 + |𝑧|2 )2 (1 + |𝑧|2 )2
By Corollary 7.15, the degree of 𝐻 is equal to the integral 2𝜋𝑖 ∫ℂℙ1 Θ𝐻 . This can
be computed explicitly by integrating over the subset 𝑈0 (which is biholomorhpic
to ℂ), since ℂℙ1 ∖ 𝑈0 has measure zero. Making the change of variables 𝑧 = 𝑟𝑒𝑖𝜃 ,
𝑑𝑧 ∧ 𝑑𝑧 = −2𝑖𝑟 𝑑𝑟 ∧ 𝑑𝜃, we obtain
𝑖 𝑑𝑧 ∧ 𝑑𝑧
deg(𝐻) =
2𝜋 ∫ℂ (1 + |𝑧|2 )2
𝑖 −2𝑖𝑟
2𝜋 ∞
= 𝑑𝑟 𝑑𝜃.
2𝜋 ∫0 ∫0 (1 + 𝑟2 )2
Then a further change of variables 𝑠 = 1 + 𝑟2 yields
1 1
2𝜋 ∞
deg(𝐻) = 𝑑𝑠 𝑑𝜃 = 1.
2𝜋 ∫ ∫ 𝑠2
//
0 0

bundles. A (1, 1)-form 𝜔 on a complex manifold 𝑀 is said to be positive if it is real

The first real Chern class gives us another useful notion of positivity for line

and 𝜔(𝑣, 𝐽 𝑣) > 0 for every nonzero real tangent vector 𝑣; a negative (1, 1)-form
is defined analogously. (Note that this is a special definition for forms on complex
manifolds; it does not make sense to ask that a differential form be literally positive
in the sense of taking only positive values, because changing the sign of one of its

for a generalization to (𝑝, 𝑝)-forms.)

arguments will always convert a positive value to a negative one. See Problem 7-9

The next lemma gives some alternative characterizations of positive forms.

216 7. Connections

Lemma 7.28. Suppose 𝜔 is a real (1, 1)-form on a complex manifold 𝑀. The

following are equivalent:
(a) 𝜔 is positive.
(b) −𝑖𝜔(𝑣, 𝑣) > 0 for every nonzero 𝑣 ∈ 𝑇 ′ 𝑀.
(c) 𝜔 restricts to a positively oriented volume form on every 1-dimensional
complex submanifold of 𝑀.

Proof. Problem 7-8. □

A complex line bundle 𝐿 on a complex manifold is said to be positive if its first

real Chern class is represented by a positive (1, 1)-form. It is negative if 𝑐1ℝ (𝐿) has
a negative representative.
The next proposition shows that for Riemann surfaces, this new notion of pos-
itivity coincides with positivity of the degree.
Proposition 7.29. Let 𝑀 be a connected, compact Riemann surface and 𝐿 → 𝑀
a holomorphic line bundle. Then 𝐿 is positive if and only if deg 𝐿 > 0.

Proof. First suppose 𝐿 is positive. Then there is a positive 2-form 𝜔 representing

𝑐1 (𝐿), and Corollary 7.15 shows that that deg 𝐿 = ∫𝑀 𝜔. Since Lemma 7.28 shows
that 𝜔 is a positively oriented volume form for 𝑀, it follows that deg 𝐿 = ∫𝑀 𝜔 > 0.
Conversely, suppose deg 𝐿 > 0, and let 𝑑𝑉𝑔 be the Riemannian volume form
for some Riemannian metric on 𝑀. Since there are no nontrivial (2, 0)-forms
or (0, 2)-forms on a complex 1-manifold, 𝑑𝑉𝑔 is a form of type (1, 1). Because
𝐻 2 (𝑀; ℝ) ≅ Hom(𝐻2 (𝑀), ℝ) is 1-dimensional, it is spanned by the cohomology
class of 𝑑𝑉𝑔 . Thus the class 𝑐1ℝ (𝐿) is represented by 𝑎 𝑑𝑉𝑔 for some real constant
𝑎. The hypothesis implies 0 < deg 𝐿 = ∫𝑀 𝑎 𝑑𝑉𝑔 = 𝑎 Vol(𝑀). Thus 𝑎 > 0, which
means 𝑎 𝑑𝑉𝑔 is a positive (1, 1)-form representing 𝑐1ℝ (𝐿). □

We end the chapter with an important result about algebraic curves in ℂℙ2 .

pose 𝑀 ⊆ ℂℙ2 is a nonsingular algebraic variety defined by a homogeneous poly-

Proposition 7.30 (Degree of the Hyperplane Bundle on a Complex Curve). Sup-

nomial 𝑃 ∶ ℂ3 → ℂ of degree 𝑑. Let 𝐿 → 𝑀 be the restriction of the hyperplane

bundle 𝐻 → ℂℙ2 . Then deg 𝐿 = 𝑑.

Proof. We will show that deg(𝐿) = 𝑑 by showing that 𝐿 has a holomorphic section
with exactly 𝑑 simple zeros.
First we dispose of the easy case. If 𝑑 = 1, then 𝑀 is a projective line. After
a projective transformation, we may assume 𝑀 is the standard embedding of ℂℙ1
in ℂℙ2 , and the restriction of 𝐻 to 𝑀 is isomorphic to the hyperplane bundle of
ℂℙ1 (as can be verified by examining transition functions). Thus deg(𝐿) = 1 by
the result of Example 7.27.
The Chern Connection 217

For the case 𝑑 > 1, we start by finding a projective line Π ⊆ ℂℙ2 that intersects
𝑀 transversely. Let (ℂℙ2 )∗ denote the dual projective space (Example 2.32), and
consider the map Φ ∶ 𝑀 → (ℂℙ2 )∗ that sends a point 𝑝 ∈ 𝑀 to the projective
tangent space to 𝑀 at 𝑝 (see Prop. 2.33). This map is smooth (as can be verified
by writing it in any affine coordinates), so its image has measure zero by an easy

line Π ∈ (ℂℙ2 )∗ not in the image of Φ, which is to say that 𝑇𝑝′ Π ≠ 𝑇𝑝′ 𝑀 at each
application of Sard’s theorem [LeeSM, Cor. 6.11]. Thus there is some projective

point 𝑝 ∈ Π ∩ 𝑀. Since both holomorphic tangent spaces are 1-dimensional, their

intersection is trivial, and then by conjugation it follows that 𝑇𝑝″ Π∩𝑇𝑝″ 𝑀 = {0} for
each such 𝑝, and thus 𝑇𝑝 Π ∩ 𝑇𝑝 𝑀 = {0}. By applying a projective transformation,
we can arrange that Π is the projective line {[𝑤0 , 𝑤1 , 𝑤2 ] ∶ 𝑤2 = 0}.
Since the polynomial 𝑃 has degree 𝑑 > 1, there is some point 𝑏 = [𝑏0 , 𝑏1 , 0] that
lies on Π but not on 𝑀. The projective transformation 𝐴[𝑤0 , 𝑤1 , 𝑤2 ] = [𝑏1 𝑤0 +
𝑏0 𝑤1 , 𝑏1 𝑤1 −𝑏0 𝑤0 , 𝑤2 ] takes [0, 1, 0] to 𝑏 and leaves Π invariant, so after replacing
𝑀 by 𝐴−1 (𝑀), we can arrange that [0, 1, 0] ∉ 𝑀.
By formula (3.15), the linear function 𝑤2 ∶ ℂ3 → ℂ defines a holomorphic
section 𝜑𝑤2 of 𝐻 that vanishes exactly on Π. Let 𝜎 be the section of 𝐿 = 𝐻|𝑀
obtained by restricting 𝜑𝑤2 . Its zeros are exactly the points where 𝑀 intersects Π.
Every such point can be written in the form [𝑤0 , 𝑤1 , 0], where 𝑃 (𝑤0 , 𝑤1 , 0) = 0.
Since [0, 1, 0] is not one of those points, we can write each such point uniquely
in the form [1, 𝜁, 0], for 𝜁 a solution to the polynomial equation 𝑞(𝜁) = 0, where
𝑞(𝜁) = 𝑃 (1, 𝜁, 0). Thus all such points lie in the affine coordinate domain 𝑈0 =
{[𝑤0 , 𝑤1 , 𝑤2 ] ∶ 𝑤0 ≠ 0}.
We need to check several things: that 𝑞 is actually a polynomial of degree 𝑑;
that each zero of 𝑞 has multiplicity 1; and correspondingly that each zero of 𝜎 has
multiplicity 1. Once we have verified these facts, it will follow from the fundamen-
tal theorem of algebra that 𝑞 has exactly 𝑑 zeros, all of multiplicity 1; and therefore
𝜎 has 𝑑 zeros of multiplicity 1. Then Theorem 7.22 implies that deg(𝐿) = 𝑑.
To see that 𝑞 has degree 𝑑, note first that it has degree at most 𝑑 because it is
obtained from 𝑃 by setting some variables equal to constants. Our normalizations
have ensured that 𝑃 (0, 1, 0) ≠ 0. Since 𝑃 (0, 1, 0) is equal to the coefficient of
(𝑤1 )𝑑 in the expression for 𝑃 , it follows that this coefficient is nonzero, so 𝑞(𝜁) has
a nonzero 𝜁 𝑑 term.
To see that the zeros of 𝑞 all have multiplicity 1, assume for contradiction
that 𝑎 ∈ ℂ is a zero of multiplicity greater than 1. This implies 0 = 𝑞 ′ (𝑎) =
(𝜕𝑃 /𝜕𝑤1 )(1, 𝑎, 0). In affine coordinates (𝑧1 , 𝑧2 ) ↔ [1, 𝑧1 , 𝑧2 ] on 𝑈0 , 𝑀 ∩ 𝑈0 is
defined by the equation 𝑃 (1, 𝑧1 , 𝑧2 ) = 0, and Π ∩ 𝑈0 is defined by 𝑧2 = 0. Since
(𝜕𝑃 /𝜕𝑤1 )(1, 𝑎, 0) = 0, it follows that both 𝑇(𝑎,0)
′
𝑀 and 𝑇(𝑎,0)
′
Π are spanned in these
coordinates by 𝜕/𝜕𝑧1 , which contradicts the fact that 𝑀 and Π intersect transversely.
218 7. Connections

Finally, to see that the zeros of 𝜎 are simple, note that the restriction of 𝐻 to 𝑈0
has a global nonvanishing section 𝑠 = 𝜑𝑤0 |𝑈0 , and in terms of that section 𝜎|𝑀∩𝑈0
can be written 𝜎(𝑧1 , 𝑧2 ) = 𝑧2 𝑠. At a point (𝑎, 0) ∈ 𝑀 ∩ Π ∩ 𝑈0 , the argument above
shows that (𝜕𝑃 /𝜕𝑤1 )(1, 𝑎, 0) = 𝑞 ′ (𝑎) ≠ 0, so by the holomorphic implicit function
theorem we can solve the equation 𝑃 (1, 𝑧1 , 𝑧2 ) = 0 for 𝑧1 in some neighborhood
𝑊 of (𝑎, 0), and write 𝑀 ∩ 𝑊 = {(𝑧1 , 𝑧2 ) ∶ 𝑧1 = 𝑓 (𝑧2 )} for some holomorphic
function 𝑓 that satisfies 𝑃 (1, 𝑓 (𝑧2 ), 𝑧2 ) = 0 and 𝑓 (0) = 𝑎. This gives a local
parametrization of 𝑀 in the form (𝑧1 , 𝑧2 ) = 𝜒(𝜁) = (𝑓 (𝜁), 𝜁), with 𝜒(0) = (𝑎, 0).
Pulling back 𝜎 by this parametrization, we find 𝜒 ∗ 𝜎(𝜁) = 𝜁𝜒 ∗ 𝑠, which has a zero
of multiplicity 1 at 𝜁 = 0. This completes the proof that deg(𝐿) = 𝑑. □

Problems
7-1. Prove the converse of Proposition 7.1: if ∇ is a connection on a Hermit-
ian vector bundle whose matrix of connection 1-forms is skew-Hermitian
with respect to every local orthonormal frame, then ∇ is a metric connec-
tion.
7-2. Suppose 𝐸 → 𝑀 is a holomorphic vector bundle with a Hermitian fiber
metric ℎ, and 𝐸 ′ ⊆ 𝐸 is a holomorphic subbundle with the fiber metric
ℎ′ given by restricting ℎ. For any section 𝜎 of 𝐸, let 𝜎 ⊤ and 𝜎 ⟂ be the
orthogonal projections of 𝜎 onto 𝐸 ′ and (𝐸 ′ )⟂ , respectively. Denoting
the Chern connections on 𝐸 and 𝐸 ′ by ∇ and ∇′ , respectively, show that
∇′𝑋 𝜎 = (∇𝑋 𝜎)⊤ for every smooth vector field 𝑋 on 𝑀 and smooth section
𝜎 of 𝐸 ′ .
7-3. Let 𝐸 → 𝑀 be a smooth complex vector bundle endowed with a Hermit-
ian fiber metric ℎ.
(a) Show that ℎ determines a smooth conjugate-linear bundle isomor-
phism ̂
ℎ ∶ 𝐸 → 𝐸 ∗ by
̂
ℎ(𝜎)(𝜏) = ⟨𝜏, 𝜎⟩ℎ .
(Note the reversal of order on the right-hand side.)
(b) Show that the formula
⟨𝜑, 𝜓⟩ℎ∗ = ⟨̂
ℎ−1 (𝜓), ̂
ℎ−1 (𝜑)⟩ℎ .
defines a Hermitian fiber metric ℎ∗ on the dual bundle 𝐸 ∗ , called the

7-4. Let 𝐸 → 𝑀 be a smooth complex vector bundle, and let ∇ be a connection

dual metric.

on 𝐸. Define a map ∇∗ ∶ Γ(𝑇ℂ 𝑀) × Γ(𝐸 ∗ ) → Γ(𝐸 ∗ ) by

(∇∗𝑋 𝜑)(𝜎) = 𝑋(𝜑(𝜎)) − 𝜑(∇𝑋 𝜎)
for 𝑋 ∈ Γ(𝑇ℂ 𝑀), 𝜑 ∈ Γ(𝐸 ∗ ), and 𝜎 ∈ Γ(𝐸).
(a) Show that ∇∗ is a connection on 𝐸 ∗ , called the dual connection.
Problems 219

(b) Now suppose 𝐸 is endowed with a Hermitian fiber metric ℎ, and let
ℎ∗ be the dual metric on 𝐸 ∗ (Problem 7-3). Show that if ∇ is a metric
connection, then

∇∗𝑋 (̂
ℎ(𝜎)) = ̂
ℎ(∇𝑋 𝜎 ),

and conclude that ∇∗ is also a metric connection.

(c) Show that if 𝐸 is a holomorphic Hermitian vector bundle and ∇ is its
Chern connection, then ∇∗ is the Chern connection of 𝐸 ∗ .
7-5. Let 𝐸 → 𝑀 be a smooth complex vector bundle, let ∇ be a connection on
𝐸, and let ∇∗ be the dual connection on 𝐸 ∗ (Problem 7-5). Suppose (𝑠𝑗 )
is a holomorphic local frame for 𝐸, and 𝜃𝑘 and Θ𝑘 are its connection and
𝑗 𝑗

curvature forms, respectively. Let (𝜀𝑗 ) be the dual frame for 𝐸 ∗ defined by
𝜀𝑗 (𝑠𝑘 ) = 𝛿𝑗𝑘 , and let 𝜃𝑗∗𝑘 and Θ∗𝑘
𝑗 be the connection and curvature forms
of ∇∗ , satisfying

∇∗𝑋 𝜀𝑗 = 𝜃𝑗∗𝑘 (𝑋)𝜀𝑘 , Θ∗ (𝑋, 𝑌 )𝜀𝑗 = Θ∗𝑘

𝑗 (𝑋, 𝑌 )𝜀𝑘 .

to ensure that the indices of the connection and curvature forms for ∇∗
(We write the dual frame with lower indices instead of upper ones so as

function the same as those for ∇.) Show that

𝜃𝑗∗𝑘 = −𝜃𝑘 , Θ∗𝑘

𝑗 = −Θ𝑘 .
𝑗 𝑗

7-6. Let 𝐸, 𝐸 ′ → 𝑀 be smooth complex vector bundles endowed with con-

nections ∇ and ∇′ .
(a) Show that there is a unique connection ∇⊗ on 𝐸 ⊗ 𝐸 ′ , called the
tensor product connection, that satisfies ∇⊗𝑋 (𝜎 ⊗ 𝜎 ) = ∇𝑋 𝜎 ⊗ 𝜎 +
′ ′

𝜎 ⊗ ∇𝑋 𝜎 for all 𝜎 ∈ Γ(𝐸) and 𝜎 ∈ Γ(𝐸 ).

′ ′ ′ ′

(b) Show that the curvature of ∇⊗ satisfies

Θ⊗ (𝑋, 𝑌 )(𝜎 ⊗ 𝜏) = (Θ(𝑋, 𝑌 )𝜎 ) ⊗ 𝜏 + 𝜎 ⊗ (Θ′ (𝑋, 𝑌 )𝜏 ).

(c) Show that if 𝐸 and 𝐸 ′ are holomorphic Hermitian bundles and ∇ and
∇′ are their Chern connections, then ∇⊗ is the Chern connection on
𝐸 ⊗ 𝐸 ′.
7-7. Let ∇ be a connection on a smooth vector bundle 𝐸 → 𝑀 and Θ
its curvature. Prove the differential Bianchi identity 𝐷Θ = 0, where
𝐷 ∶ ℰ 2 (End 𝐸) → ℰ 3 (End 𝐸) is the exterior covariant derivative opera-
tor associated with the tensor product connection on End 𝐸 ≅ 𝐸 ⊗ 𝐸 ∗ .
7-8. Prove Lemma 7.28 (characterizations of positive (1, 1)-forms).
220 7. Connections

7-9. Let 𝑀 be a complex 𝑛-manifold. Define a positive (𝒑, 𝒑)-form on 𝑀 to

be a real (𝑝, 𝑝)-form that restricts to a positively oriented volume form on
every 𝑝-dimensional complex submanifold of 𝑀.
(a) Show that a real (𝑝, 𝑝)-form 𝜔 is positive if and only if
(−𝑖)𝑝 𝜔(𝑣1 , 𝑣1 , … , 𝑣𝑝 , 𝑣𝑝 ) > 0
whenever (𝑣1 , … , 𝑣𝑝 ) is a linearly independent 𝑝-tuple of elements of
𝑇𝑥′ 𝑀 for 𝑥 ∈ 𝑀.
(b) Show that if 𝜑 is a nonzero (𝑛, 0)-form, then 𝑖𝑛 𝜑 ∧ 𝜑 is positive.
2

7-10. Let 𝑀 be a complex 𝑛-manifold, and let 𝜋 ∶ 𝐸 → 𝑀 be a smooth com-

plex vector bundle. Given a connection ∇ on 𝐸, let 𝐷 be the associ-
ated exterior covariant derivative operator, and for each 𝑞 > 0 define
an operator 𝐷″ ∶ ℰ 0,𝑞 (𝑀; 𝐸) → ℰ 0,𝑞+1 (𝑀; 𝐸) by 𝐷″ = 𝜋 0,𝑞+1 ∘ 𝐷.
Show that if there is a connection on 𝐸 that satisfies 𝐷″ ∘ 𝐷″ = 0 on
ℰ 0,0 (𝑀; 𝐸), then 𝐸 has a unique structure as a holomorphic vector bun-
dle such that the holomorphic sections of 𝐸 are exactly those in the kernel
of 𝐷″ . [Hint: Show that for any smooth local frame (𝑠1 , … , 𝑠𝑚 ) for 𝐸
over 𝑈 ⊆ 𝑀, the (0, 1)-forms 𝜃𝑘 on 𝑈 defined by 𝐷″ (𝑠𝑘 ) = 𝜃𝑘 ⊗ 𝑠𝑗
𝑗 𝑗

satisfy 𝜕𝜃𝑘 + 𝜃𝑙 ∧ 𝜃𝑘𝑙 = 0. Let (𝑧𝑗 ) be local holomorphic coordinates for

𝑗 𝑗

𝑈 and let (𝑧, 𝑏) = (𝑧1 , … , 𝑧𝑛 , 𝑏1 , … , 𝑏𝑚 ) be the (complex-valued) coordi-

nates on the open set 𝜋 −1 (𝑈 ) ⊆ 𝐸 defined by the local frame (𝑠𝑘 ), via the
correspondence (𝑧, 𝑏) ↔ 𝑏𝑘 𝑠𝑘 (𝑧). Show that there is a unique integrable
almost complex structure on the total space of 𝐸 such that Λ1,0 (𝐸) is lo-
cally spanned by {𝜋 ∗ 𝑑𝑧𝑗 , 𝑑𝑏𝑙 + 𝑏𝑘 𝜋 ∗ 𝜃𝑘𝑙 ∶ 𝑗 = 1, … , 𝑛, 𝑙 = 1, … , 𝑚},
and apply the Newlander–Nirenberg theorem.]
 10.1090/gsm/244/08

Chapter 8

Hermitian and Kähler

Manifolds

In this chapter, we explore the interplay between holomorphic structures and met-
rics on the tangent bundle. We begin by discussing basic properties of Hermit-
ian metrics on the tangent bundle and their relationships with Riemannian metrics.
Then for the rest of the chapter we focus on the special case of Kähler metrics, which
are Hermitian metrics that satisfy an additional condition ensuring a closer relation-
ship between the Riemannian structure and the holomorphic structure. Complex
manifolds that admit Kähler metrics are far and away the most important class of
complex manifolds.

Hermitian Metrics on the Tangent Bundle

Let 𝑀 be a complex manifold. Because the almost complex structure 𝐽 turns the
tangent bundle into a complex vector bundle 𝑇𝐽 𝑀, we can look for Hermitian fiber
metrics on 𝑇𝐽 𝑀. Because 𝐽 plays the role of multiplication by 𝑖 on 𝑇𝐽 𝑀, a Her-
mitian fiber metric in this case is a map ℎ ∶ Γ(𝑇𝐽 𝑀) × Γ(𝑇𝐽 𝑀) → 𝐶 ∞ (𝑀; ℂ) that
satisfies all of the following:

• ℎ is bilinear over 𝐶 ∞ (𝑀; ℝ);

• ℎ(𝐽 𝑋, 𝑌 ) = 𝑖ℎ(𝑋, 𝑌 );
• ℎ(𝑋, 𝐽 𝑌 ) = −𝑖ℎ(𝑋, 𝑌 );
• ℎ(𝑌 , 𝑋) = ℎ(𝑋, 𝑌 );
• ℎ(𝑋, 𝑋) > 0 at points where 𝑋 ≠ 0.

221
222 8. Hermitian and Kähler Manifolds

A Hermitian fiber metric is not a Riemannian metric, because it cannot take on

only real values; but the next lemma shows that its real part is.
Lemma 8.1. Suppose 𝑀 is a complex manifold and ℎ is a Hermitian fiber metric
on 𝑇𝐽 𝑀. Then 𝑔 = Re ℎ is a Riemannian metric on 𝑀.

Proof. It follows directly from the definition that 𝑔 is smooth, positive definite, and
bilinear over 𝐶 ∞ (𝑀; ℝ). It remains only to show that it is symmetric. We compute

𝑔(𝑋, 𝑌 ) = 21 (ℎ(𝑋, 𝑌 ) + ℎ(𝑋, 𝑌 )) = 21 (ℎ(𝑋, 𝑌 ) + ℎ(𝑌 , 𝑋)),

which is unchanged when 𝑋 and 𝑌 are swapped. □

It is natural to ask what can be said about the imaginary part. The next lemma
answers that question.
Lemma 8.2. Suppose 𝑀 is a complex manifold and ℎ is a Hermitian fiber metric
on 𝑇𝐽 𝑀. Then 𝜔 = − Im ℎ is a 2-form of type (1, 1).

Proof. As before, it is immediate that 𝜔 is smooth and bilinear over 𝐶 ∞ (𝑀; ℝ),
so we need only show that it is antisymmetric and of type (1, 1). Antisymmetry is
another simple computation:
1 1
𝜔(𝑋, 𝑌 ) = − ℎ(𝑋, 𝑌 ) − ℎ(𝑋, 𝑌 )) = − (ℎ(𝑋, 𝑌 ) − ℎ(𝑌 , 𝑋)),
2𝑖 ( 2𝑖
and this last expression changes sign when 𝑋 and 𝑌 are swapped.
To see that 𝜔 is of type (1, 1), note that the properties of ℎ guarantee that
ℎ(𝐽 𝑋, 𝐽 𝑌 ) = (𝑖)(−𝑖)ℎ(𝑋, 𝑌 ) = ℎ(𝑋, 𝑌 ) for all real vector fields 𝑋 and 𝑌 . It
follows that

𝜔(𝐽 𝑋, 𝐽 𝑌 ) = − Im ℎ(𝐽 𝑋, 𝐽 𝑌 ) = − Im ℎ(𝑋, 𝑌 ) = 𝜔(𝑋, 𝑌 ),

so 𝜔 is of type (1, 1) by the result of Problem 4-1. □

The reasons for the choice of a negative sign in the definition of 𝜔 will become

Another natural question to ask is this: Given a Riemannian metric 𝑔 on a

clear below (see Example 8.6 and Proposition 8.9).

complex manifold 𝑀, can we find a 2-form 𝜔 such that ℎ = 𝑔 − 𝑖𝜔 is a Hermitian

fiber metric on 𝑇𝐽 𝑀? The first thing to note is that if there is such a form 𝜔, it is
uniquely determined by 𝑔.
Lemma 8.3. Suppose 𝑀 is a complex manifold and ℎ is a Hermitian fiber metric
on 𝑇𝐽 𝑀. Let 𝑔 = Re ℎ and 𝜔 = − Im ℎ. Then

(8.1) 𝜔(𝑋, 𝑌 ) = 𝑔(𝐽 𝑋, 𝑌 ) for all 𝑋, 𝑌 ∈ Γ(𝑇 𝑀).

Hermitian Metrics on the Tangent Bundle 223

1
Proof. Once again, a computation:

𝑔(𝐽 𝑋, 𝑌 ) = (ℎ(𝐽 𝑋, 𝑌 ) + ℎ(𝐽 𝑋, 𝑌 ))

2
1
= (𝑖ℎ(𝑋, 𝑌 ) + 𝑖ℎ(𝑋, 𝑌 ))
2
𝑖
= (ℎ(𝑋, 𝑌 ) − ℎ(𝑋, 𝑌 ))
2
= 𝜔(𝑋, 𝑌 ). □

So now the question becomes: Given a Riemannian metric 𝑔, if we define a

2-form 𝜔 by (8.1), when is ℎ = 𝑔 − 𝑖𝜔 a Hermitian fiber metric on 𝑇𝐽 𝑀? The next
proposition answers that question in several ways.
Proposition 8.4. Let 𝑀 be a complex manifold 𝑀 and 𝐽 ∶ 𝑇 𝑀 → 𝑇 𝑀 its asso-
ciated almost complex structure. Given a Riemannian metric 𝑔 on 𝑀, define 𝜔, ℎ,
and 𝑔ℂ by
• 𝜔(𝑋, 𝑌 ) = 𝑔(𝐽 𝑋, 𝑌 ),
• ℎ = 𝑔 − 𝑖𝜔,
• 𝑔ℂ is the extension of 𝑔 by complex bilinearity to act on pairs of complex
vector fields.
The following are equivalent.
(a) ℎ is a Hermitian fiber metric on 𝑇𝐽 𝑀.
(b) 𝐽 is an orthogonal map with respect to 𝑔.
(c) 𝐽 is a skew-symmetric map with respect to 𝑔.
(d) 𝑔ℂ (𝑋, 𝑌 ) = 0 if 𝑋 and 𝑌 are both sections of 𝑇 ′ 𝑀.
(e) 𝜔 is skew-symmetric.

Proof. We will prove (a) ⇒ (b) ⇒ (c) ⇒ (d) ⇒ (e) ⇒ (a).

(a) ⇒ (b): Suppose ℎ is a Hermitian fiber metric. Because ℎ is sesquilinear, it
satisfies ℎ(𝐽 𝑋, 𝐽 𝑌 ) = (𝑖)(−𝑖)ℎ(𝑋, 𝑌 ) = ℎ(𝑋, 𝑌 ). Thus
𝑔(𝐽 𝑋, 𝐽 𝑌 ) = 21 (ℎ(𝐽 𝑋, 𝐽 𝑌 ) + ℎ(𝐽 𝑋, 𝐽 𝑌 ))
= 21 (ℎ(𝑋, 𝑌 ) + ℎ(𝑋, 𝑌 )) = 𝑔(𝑋, 𝑌 ).

(b) ⇒ (c): If 𝐽 is orthogonal, then 𝑔(𝐽 𝑋, 𝑌 ) = 𝑔(𝐽 2 𝑋, 𝐽 𝑌 ) = −𝑔(𝑋, 𝐽 𝑌 ).

(c) ⇒ (d): Assuming 𝐽 is skew-symmetric, its complexification 𝐽 ∶ 𝑇ℂ 𝑀 →
𝑇ℂ 𝑀 is also skew-symmetric with respect to 𝑔ℂ . Suppose 𝑍, 𝑊 ∈ Γ(𝑇 ′ 𝑀). Then
𝑖𝑔ℂ (𝑍, 𝑊 ) = 𝑔ℂ (𝑖𝑍, 𝑊 ) = 𝑔ℂ (𝐽 𝑍, 𝑊 )
= −𝑔ℂ (𝑍, 𝐽 𝑊 ) = −𝑔ℂ (𝑍, 𝑖𝑊 ) = −𝑖𝑔ℂ (𝑍, 𝑊 ),
from which it follows that 𝑔ℂ (𝑍, 𝑊 ) = 0.
224 8. Hermitian and Kähler Manifolds

(d) ⇒ (e); Suppose 𝑔ℂ vanishes when applied to pairs of sections of 𝑇 ′ 𝑀.

Because 𝑔ℂ is the complexification of a real tensor field, by conjugation it also
vanishes when applied to pairs of sections of 𝑇 ″ 𝑀. Given 𝑋, 𝑌 ∈ Γ(𝑇 𝑀), we can
use the decomposition 𝑇ℂ 𝑀 = 𝑇 ′ 𝑀 ⊕𝑇 ″ 𝑀 to write 𝑋 = 𝑍 +𝑍 and 𝑌 = 𝑊 +𝑊 ,
with 𝑍, 𝑊 ∈ Γ(𝑇 ′ 𝑀). Then

𝜔(𝑋, 𝑌 ) = 𝑔(𝐽 𝑋, 𝑌 ) = 𝑔ℂ (𝐽 (𝑍 + 𝑍), 𝑊 + 𝑊 )

= 𝑔ℂ (𝑖𝑍 − 𝑖𝑍, 𝑊 + 𝑊 )
= 𝑖(𝑔ℂ (𝑍, 𝑊 ) + 𝑔ℂ (𝑍, 𝑊 ) − 𝑔ℂ (𝑍, 𝑊 ) − 𝑔ℂ (𝑍, 𝑊 ))
= 𝑖(𝑔ℂ (𝑍, 𝑊 ) − 𝑔ℂ (𝑍, 𝑊 )).

Interchanging 𝑋 and 𝑌 has the effect of interchanging 𝑍 and 𝑊 , and this last

(e) ⇒ (a): Now suppose 𝜔 is skew-symmetric. Then ℎ = 𝑔 − 𝑖𝜔 is certainly

expression manifestly changes sign under that interchange.

smooth and bilinear over 𝐶 ∞ (𝑀, ℝ), and it is positive definite because the skew
symmetry of 𝜔 implies

ℎ(𝑋, 𝑋) = 𝑔(𝑋, 𝑋) − 𝑖𝜔(𝑋, 𝑋) = 𝑔(𝑋, 𝑋) > 0, where 𝑋 ≠ 0.

To see that ℎ is conjugate symmetric, compute

ℎ(𝑋, 𝑌 ) − ℎ(𝑌 , 𝑋) = 𝑔(𝑋, 𝑌 ) − 𝑖𝜔(𝑋, 𝑌 ) − (𝑔(𝑌 , 𝑋) − 𝑖𝜔(𝑌 , 𝑋))

= 𝑔(𝑋, 𝑌 ) − 𝑖𝜔(𝑋, 𝑌 ) − 𝑔(𝑌 , 𝑋) − 𝑖𝜔(𝑌 , 𝑋)
= 0.

And for complex linearity in the first variable, we have

ℎ(𝐽 𝑋, 𝑌 ) = 𝑔(𝐽 𝑋, 𝑌 ) − 𝑖𝜔(𝐽 𝑋, 𝑌 ) = 𝜔(𝑋, 𝑌 ) − 𝑖𝑔(𝐽 2 𝑋, 𝑌 )

= 𝜔(𝑋, 𝑌 ) + 𝑖𝑔(𝑋, 𝑌 ) = 𝑖ℎ(𝑋, 𝑌 ).

Finally, conjugate linearity in the second variable follows from the two previous
computations:

ℎ(𝑋, 𝐽 𝑌 ) = ℎ(𝐽 𝑌 , 𝑋) = 𝑖ℎ(𝑌 , 𝑋) = −𝑖ℎ(𝑋, 𝑌 ). □

A Hermitian metric on a complex manifold 𝑀 is a Riemannian metric for

which 𝐽 is orthogonal, and a complex manifold endowed with a Hermitian structure
is called a Hermitian manifold. Given a Hermitian manifold (𝑀, 𝑔), the 2-form
𝜔 = 𝑔(𝐽 ⋅, ⋅) is called the fundamental 2-form of the Hermitian metric. (Although
this terminology is quite common, it is important to remember that the metric 𝑔
itself is not a Hermitian fiber metric on 𝑇𝐽 𝑀; that role is played by ℎ = 𝑔 − 𝑖𝜔.)
Lemma 8.5. Every complex manifold has a Hermitian metric.
Hermitian Metrics on the Tangent Bundle 225

Proof. Given a complex manifold 𝑀, let 𝑔0 be an arbitrary Riemannian metric on

𝑀, and define another Riemannian metric 𝑔 by
𝑔(𝑋, 𝑌 ) = 𝑔0 (𝑋, 𝑌 ) + 𝑔0 (𝐽 𝑋, 𝐽 𝑌 ).
It follows easily that 𝑔(𝐽 𝑋, 𝐽 𝑌 ) = 𝑔(𝑋, 𝑌 ). □

Example 8.6 (The Standard Metric on ℂ𝒏 ). On ℂ𝑛 with standard holomorphic

coordinates 𝑧𝑗 = 𝑥𝑗 + 𝑖𝑦𝑗 , we have the Euclidean metric
𝑛
𝑔E = (𝑑𝑥𝑗 )2 + (𝑑𝑦𝑗 )2 .
∑
𝑗=1

𝜕 𝜕 𝜕 𝜕
𝐽 = 𝑗 and 𝐽 𝑗 = − 𝑗 ,
Because

𝜕𝑥 𝑗 𝜕𝑦 𝜕𝑦 𝜕𝑥
𝐽 takes an orthonormal frame to an orthonormal frame, so it is orthogonal. Thus
𝑔E is Hermitian. Its fundamental 2-form 𝜔E satisfies
𝜕 𝜕 𝜕 𝜕
𝜔E ( 𝑗 , 𝑘 ) = 𝑔E ( 𝑗 , 𝑘 ) = 0,
𝜕𝑥 𝜕𝑥 𝜕𝑦 𝜕𝑥
𝜕 𝜕 𝜕 𝜕
𝜔E ( 𝑗 , 𝑘 ) = 𝑔E ( 𝑗 , 𝑘 ) = 𝛿𝑗𝑘 ,
𝜕𝑥 𝜕𝑦 𝜕𝑦 𝜕𝑦
𝜕 𝜕 𝜕 𝜕
𝜔E ( 𝑗 , 𝑘 ) = −𝑔E ( 𝑗 , 𝑘 ) = −𝛿𝑗𝑘 ,
𝜕𝑦 𝜕𝑥 𝜕𝑥 𝜕𝑥
𝜕 𝜕 𝜕 𝜕
𝜔E ( 𝑗 , 𝑘 ) = −𝑔E ( 𝑗 , 𝑘 ) = 0.
𝜕𝑦 𝜕𝑦 𝜕𝑥 𝜕𝑦
It follows that 𝜔E has the coordinate expression
𝑛
𝜔E = 𝑑𝑥𝑗 ∧ 𝑑𝑦𝑗 .
∑
(8.2)
𝑗=1

This 2-form is called the standard symplectic form on ℝ2𝒏 (see [LeeSM, Example

the fundamental 2-form.

22.9(a)]). This formula is one reason for the choice of the sign in our definition of
//
In the discussion so far, we have assumed that we started with a Riemannian

on 𝑇𝐽 𝑀. We can also start with a 2-form 𝜔 and ask the analogous question: Is there
metric and looked for conditions under which it determines a Hermitian fiber metric

a Hermitian metric 𝑔 for which 𝜔 is the fundamental 2-form? The next lemma
shows that such a 𝑔, if it exists, is determined by 𝜔.
Lemma 8.7. Suppose 𝑔 is a Hermitian metric on a complex manifold 𝑀 and 𝜔 is
its fundamental 2-form. Then
𝑔(𝑋, 𝑌 ) = 𝜔(𝑋, 𝐽 𝑌 ) for all 𝑋, 𝑌 ∈ Γ(𝑇 𝑀).

► Exercise 8.8. Prove this lemma.

226 8. Hermitian and Kähler Manifolds

Thus another appropriate question is: Given a 2-form 𝜔, if we define 𝑔(𝑋, 𝑌 ) =

𝜔(𝑋, 𝐽 𝑌 ), when is 𝑔 a Hermitian metric? Lemma 8.2 shows that one necessary
condition is that 𝜔 must be of type (1, 1). Another obvious necessary condition
is that 𝜔 must be a positive form, because 𝑔 must be positive definite. The next
proposition shows that these two conditions are also sufficient.
Proposition 8.9. Suppose 𝑀 is a complex manifold and 𝜔 is a smooth 2-form on
𝑀. Define a 2-tensor 𝑔 by 𝑔(𝑋, 𝑌 ) = 𝜔(𝑋, 𝐽 𝑌 ). Then 𝑔 is a Hermitian metric
with 𝜔 as its fundamental 2-form if and only if 𝜔 is a positive (1, 1)-form.

Proof. If 𝑔 is a Hermitian metric with 𝜔 as its fundamental 2-form, then 𝜔 is a

positive (1, 1)-form as noted above.
Conversely, suppose 𝜔 is a smooth positive (1, 1)-form. Problem 4-1 shows
that 𝜔(𝐽 𝑋, 𝐽 𝑌 ) = 𝜔(𝑋, 𝑌 ) for all vector fields 𝑋, 𝑌 . Let 𝑔(𝑋, 𝑌 ) = 𝜔(𝑋, 𝐽 𝑌 ),
and note that 𝑔 is smooth and bilinear over 𝐶 ∞ (𝑀), and positivity of 𝜔 ensures that
𝑔 is positive definite. To ensure that 𝑔 is a Riemannian metric, we need only check
that it is symmetric. Using the observation above, we find
𝑔(𝑋, 𝑌 ) = 𝜔(𝑋, 𝐽 𝑌 ) = 𝜔(𝐽 𝑋, 𝐽 2 𝑌 ) = −𝜔(𝐽 𝑋, 𝑌 ) = 𝜔(𝑌 , 𝐽 𝑋) = 𝑔(𝑌 , 𝑋).
To finish the proof, we use the fact that 𝑔(𝐽 𝑋, 𝑌 ) = 𝜔(𝐽 𝑋, 𝐽 𝑌 ) = 𝜔(𝑋, 𝑌 ),
so 𝜔 is the fundamental 2-form associated with 𝑔; and since 𝜔 is skew-symmetric,
Proposition 8.4 shows that 𝑔 is Hermitian. □

Since positivity of a (1, 1)-form is a natural condition because of its relation-

ship with orientations of 1-dimensional complex submanifolds (Lemma 7.28), the

definition of the fundamental 2-form.

preceding lemma gives another justification for the choice of the minus sign in the

morphism 𝑇𝐽 𝑀 ≅ 𝑇 ′ 𝑀 (Prop. 1.58), the bundle 𝑇𝐽 𝑀 inherits the structure of a

There is one more natural question we might ask: By virtue of the natural iso-

holomorphic vector bundle. Thus a Hermitian fiber metric ℎ = 𝑔 − 𝑖𝜔 determines

a Chern connection on 𝑇 𝑀. One might wonder whether the Chern connection of
ℎ coincides with the Levi-Civita connection of 𝑔. The answer is “not always,” and
we will explore the conditions under which they do coincide in the next section.

Kähler Metrics

mental 2-form 𝜔 is closed. A complex manifold equipped with a Kähler metric is

A Hermitian metric on a complex manifold is called a Kähler metric if its funda-

called a Kähler manifold, This seemingly innocuous condition turns out to have

Because every Hermitian metric is completely determined by its fundamental 2-

deep and wide-ranging consequences.

complex manifold is a smooth, closed, positive (1, 1)-form. By virtue of Proposition

form, it is frequently easiest to define a Kähler metric that way. A Kähler form on a
Kähler Metrics 227

8.9, every Kähler form 𝜔 determines a Kähler metric via 𝑔 = 𝜔(⋅, 𝐽 ⋅). A Kähler
form is, in particular, a symplectic form, which is a smooth closed real 2-form that
is nondegenerate in the sense that the map from 𝑇 𝑀 to 𝑇 ∗ 𝑀 given by 𝑣 ↦ 𝑣 ⅃ 𝜔
is injective (see [LeeSM, Chap. 22]).

[𝜔] ∈ 𝐻dR
2
(𝑀; ℝ), called the Kähler class of the given metric.
Because a Kähler form is closed and real, it determines a real cohomology class

some additional index conventions. Suppose 𝑀 is an 𝑛-dimensional complex man-

In order to perform computations with Kähler metrics efficiently, we introduce

ifold. Given any local holomorphic coordinates (𝑧1 , … , 𝑧𝑛 ) for 𝑀, we will gen-
erally use the local frame (𝜕/𝜕𝑧1 , … , 𝜕/𝜕𝑧𝑛 , 𝜕/𝜕𝑧1 , … , 𝜕/𝜕𝑧𝑛 ) for the complexified
tangent bundle, numbered from 1 to 2𝑛. For each index 𝑗 ∈ {1, … , 𝑛}, we interpret
𝑗 to mean 𝑗 + 𝑛, and 𝑧𝑗 to be synonymous with 𝑧𝑗 . We use the abbreviations
𝜕 𝜕
𝜕𝑗 = , 𝜕𝑗 = ,
𝜕𝑧𝑗 𝜕𝑧𝑗
with implied summations over such indices going from 1 to 𝑛 = dimℂ 𝑀. Thus,
for example, the expression 𝑉 𝑗 𝜕𝑗 + 𝑉 𝑗 𝜕𝑗 is shorthand for
𝑛 𝑛
𝜕 𝜕
𝑉 𝑗
+ 𝑉𝑗 .
∑ 𝜕𝑧𝑗 ∑ 𝜕𝑧𝑗
𝑗=1 𝑗=1

Now suppose 𝑔 is a Hermitian metric. (For the time being, we are not assum-
ing it is Kähler.) As in the previous section, we extend 𝑔 to act on pairs of complex

ferentiate notationally between 𝑔 and its complex-bilinear extension, using the same
vector fields by complex bilinearity; but for simplicity, henceforth we will not dif-

symbol 𝑔 for both.

In local holomorphic coordinates, we can write the complexification of 𝑔 as

𝑔 = 𝑔𝑗𝑘 𝑑𝑧𝑗 ⊗ 𝑑𝑧𝑘 + 𝑔𝑗𝑘 𝑑𝑧𝑗 ⊗ 𝑑𝑧𝑘 + 𝑔𝑗𝑘 𝑑𝑧𝑗 ⊗ 𝑑𝑧𝑘 + 𝑔𝑗𝑘 𝑑𝑧𝑗 ⊗ 𝑑𝑧𝑘 ,

where 𝑔𝑗𝑘 = 𝑔(𝜕𝑗 , 𝜕𝑘 ), 𝑔𝑗𝑘 = 𝑔(𝜕𝑗 , 𝜕𝑘 ), etc. Because 𝑔 is the complexification of a

real symmetric tensor, we have

𝑔𝑗𝑘 = 𝑔𝑗𝑘 = 𝑔𝑘𝑗 and 𝑔𝑗𝑘 = 𝑔𝑗𝑘 = 𝑔𝑘𝑗 .

Moreover, because 𝑔 vanishes when applied to pairs of vectors in 𝑇 ′ 𝑀, we have

𝑔𝑗𝑘 = 0 and then also 𝑔𝑗𝑘 = 0 by conjugation. Thus we can write

𝑔 = 𝑔𝑗𝑘 𝑑𝑧𝑗 ⊗ 𝑑𝑧𝑘 + 𝑔𝑗𝑘 𝑑𝑧𝑗 ⊗ 𝑑𝑧𝑘

𝑑𝑧𝑗 ⊗ 𝑑𝑧𝑘 + 𝑑𝑧𝑘 ⊗ 𝑑𝑧𝑗
= 2𝑔𝑗𝑘 ( )
2
(8.3)

= 2𝑔𝑗𝑘 𝑑𝑧 𝑑𝑧 ,
𝑗 𝑘
228 8. Hermitian and Kähler Manifolds

where the juxtaposition 𝑑𝑧𝑗 𝑑𝑧𝑘 represents the symmetric product. The Riemann-
ian metric 𝑔 is then the restriction of this tensor to pairs of real vector fields. For
example, if 𝑋 and 𝑌 are real vector fields, written locally as 𝑋 = 𝑋 𝑗 𝜕𝑗 + 𝑋 𝑗 𝜕𝑗 and
𝑌 = 𝑌 𝑗 𝜕𝑗 + 𝑌 𝑗 𝜕𝑗 , then

𝑔(𝑋, 𝑌 ) = 𝑔𝑗𝑘 (𝑋 𝑗 𝑌 𝑘 + 𝑌 𝑗 𝑋 𝑘 ),

and 𝑔(𝑋, 𝑋) = 2𝑔𝑗𝑘 𝑋 𝑗 𝑋 𝑘 .

whole complexified tangent bundle. For complex vector fields 𝑋, 𝑌 , we define

For many computations, we are going to need a Hermitian fiber metric on the

(8.4) ⟨𝑋, 𝑌 ⟩ = 𝑔(𝑋, 𝑌 ),

and denote the associated norm by |𝑋| = ⟨𝑋, 𝑋⟩1/2 . It is easy to check that this
is Hermitian, and it coincides with 𝑔 when applied to real vector fields. Moreover,
because 𝑔(𝑍, 𝑊 ) = 0 when 𝑍 and 𝑊 are both sections of 𝑇 ′ 𝑀 or 𝑇 ″ 𝑀, this fiber
metric makes 𝑇 ′ 𝑀 orthogonal to 𝑇 ″ 𝑀.
Of course, we have already defined a Hermitian fiber metric on 𝑇𝐽 𝑀 by ℎ = 𝑔−
𝑖𝜔. One might hope that this new inner product matches ℎ under the isomorphism
𝜉 ∶ 𝑇𝐽 𝑀 → 𝑇 ′ 𝑀 given by 𝜉(𝑣) = 𝑣−𝑖𝐽 𝑣. Unfortunately not: for real vector fields
𝑋 and 𝑌 ,

⟨𝜉(𝑋), 𝜉(𝑌 )⟩ = ⟨𝑋 − 𝑖𝐽 𝑋, 𝑌 − 𝑖𝐽 𝑌 ⟩ = 𝑔(𝑋 − 𝑖𝐽 𝑋, 𝑌 + 𝑖𝐽 𝑌 )

= 𝑔(𝑋, 𝑌 ) − 𝑖𝑔(𝐽 𝑋, 𝑌 ) + 𝑖𝑔(𝑋, 𝐽 𝑌 ) + 𝑔(𝐽 𝑋, 𝐽 𝑌 )
= 2𝑔(𝑋, 𝑌 ) − 2𝑖𝜔(𝑋, 𝑌 ) = 2ℎ(𝑋, 𝑌 ).

We could have avoided this discrepancy by adding a factor of 1/√2 to the definition
of 𝜉; but since we will be computing norms using exclusively the inner product
⟨⋅, ⋅⟩, it is not worth the additional complication this would have brought to our

In holomorphic coordinates, for sections 𝑍, 𝑊 of 𝑇 ′ 𝑀, we have

formulas.

⟨𝑍, 𝑊 ⟩ = 𝑔 (𝑍 𝑗 𝜕𝑗 , 𝑊 𝑘 𝜕𝑘 ) = 𝑔𝑗𝑘 𝑍 𝑗 𝑊 𝑘 ,

where we have written 𝑊 𝑘 = 𝑊 𝑘 . On the other hand, if 𝑋 and 𝑌 are real, decom-
posed as 𝑋 = 𝑍 + 𝑍 and 𝑌 = 𝑊 + 𝑊 with 𝑍, 𝑊 ∈ Γ(𝑇 ′ 𝑀), then

⟨𝑋, 𝑌 ⟩ = 𝑔 (𝑍 + 𝑍, 𝑊 + 𝑊 ) = 𝑔(𝑍, 𝑊 ) + 𝑔(𝑊 , 𝑍) = 𝑔𝑗𝑘 (𝑍 𝑗 𝑊 𝑘 + 𝑊 𝑗 𝑍 𝑘 ).

In particular, for 𝑋 = 𝑍 + 𝑍, this yields

(8.5) |𝑋|2 = |𝑍 + 𝑍|2 = |𝑍|2 + |𝑍|2 = 2|𝑍|2 = 2𝑔𝑗𝑘 𝑍 𝑗 𝑍 𝑘 .

Kähler Metrics 229

We use the complexified metric 𝑔 (not the Hermitian fiber metric ⟨⋅, ⋅⟩) to define
the musical isomorphisms ♭ ∶ 𝑇ℂ 𝑀 → 𝑇ℂ∗ 𝑀 and ♯ ∶ 𝑇ℂ∗ 𝑀 → 𝑇ℂ 𝑀, called the
flat and sharp operators: for a complex vector field 𝑋, the 1-form 𝑋 ♭ is defined by
𝑋 ♭ (𝑌 ) = 𝑔(𝑋, 𝑌 ) for all complex vector fields 𝑌 ;
and ♯ is the inverse of ♭. Thus both ♯ and ♭ are smooth (but not holomorphic)
complex-linear bundle isomorphisms. For example, for 𝑍 ∈ Γ(𝑇 ′ 𝑀) and 𝑌 any
complex vector field, we have
𝑍 ♭ (𝑌 ) = 𝑔(𝑍, 𝑌 ) = 𝑔 (𝑍 𝑗 𝜕𝑗 , 𝑌 𝑘 𝜕𝑘 + 𝑌 𝑘 𝜕𝑘 ) = 𝑔𝑗𝑘 𝑍 𝑗 𝑌 𝑘 .
Thus we can write the coordinate expression of 𝑍 ♭ as 𝑍𝑘 𝑑𝑧𝑘 , where 𝑍𝑘 = 𝑔𝑗𝑘 𝑍 𝑗 .
Note that the flat operator maps 𝑇 ′ 𝑀 to Λ0,1 𝑀 and 𝑇 ″ 𝑀 to Λ1,0 𝑀. More gener-
ally, we use the matrix 𝑔𝑗𝑘 and its inverse 𝑔 𝑗𝑘 to raise and lower indices on complex

Next let us look at the coordinate expression for the fundamental 2-form 𝜔.
tensors of any type.

Since 𝜔 is a (1, 1)-form, its only nonzero terms in holomorphic coordinates are
those involving 𝑑𝑧𝑗 ∧𝑑𝑧𝑘 or 𝑑𝑧𝑗 ∧𝑑𝑧𝑘 , and by antisymmetry we can combine those
together after suitably renaming the indices. Thus we can write 𝜔 = 𝜔𝑗𝑘 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑘
for some coefficient functions 𝜔𝑗𝑘 . To determine the coefficients, we compute
𝜔𝑗𝑘 = 𝜔(𝜕𝑗 , 𝜕𝑘 ) = 𝑔(𝐽 𝜕𝑗 , 𝜕𝑘 ) = 𝑖𝑔(𝜕𝑗 , 𝜕𝑘 ) = 𝑖𝑔𝑗𝑘 .
Thus
(8.6) 𝜔 = 𝑖𝑔𝑗𝑘 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑘 .
(The reason there is a factor of 2 in formula (8.3) for 𝑔 but not in this formula for
𝜔 is because of our convention regarding wedge products—we use the convention
labeled the determinant convention in [LeeSM, p. 358], which for 1-forms 𝜔 and 𝜂
yields 𝜔∧𝜂 = 𝜔⊗𝜂−𝜂⊗𝜔, while the symmetric product is 𝜔𝜂 = 12 (𝜔 ⊗ 𝜂+𝜂 ⊗ 𝜔).

would result in a factor of 2 in the formula for 𝜔 as well.)

Using the other common convention for the wedge product, the Alt convention,

In particular, for the standard metric on ℂ𝑛 , the fundamental 2-form is given

by 𝜔 = ∑𝑗 𝑑𝑥𝑗 ∧ 𝑑𝑦𝑗 . Converting this to holomorphic coordinates using 𝑑𝑥𝑗 =
(𝑑𝑧𝑗 + 𝑑𝑧𝑗 )/2 and 𝑑𝑦𝑗 = (𝑑𝑧𝑗 − 𝑑𝑧𝑗 )/(2𝑖), we obtain
𝑖
𝜔= 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑗 , 𝑔= 𝑑𝑧𝑗 𝑑𝑧𝑗 ,
2∑ ∑
(8.7)
𝑗 𝑗

so the coefficients of the standard metric are 𝑔𝑗𝑘 = 12 𝛿𝑗𝑘 .

The next theorem illustrates many of the reasons Kähler metrics are special, and

the Hermitian ones. Before stating it, we note that because 𝑇 ′ 𝑀 is a holomorphic
at the same time gives many alternative ways to characterize Kähler metrics among

vector bundle, the isomorphism 𝜉 ∶ 𝑇𝐽 𝑀 → 𝑇 ′ 𝑀 allows us to endow 𝑇𝐽 𝑀 with

a holomorphic bundle structure as well.
230 8. Hermitian and Kähler Manifolds

Theorem 8.10 (Characterizations of Kähler Metrics). Let 𝑀 be a complex man-

ifold and 𝑔 be a Hermitian metric on 𝑀. We use the following notations:
• 𝜔 is the fundamental 2-form of 𝑔.
• ℎ = 𝑔 − 𝑖𝜔.
• ∇ is the Levi-Civita connection of 𝑔.
• ∇(ℎ) is the Chern connection of ℎ on 𝑇𝐽 𝑀.
• 𝜉 ∶ 𝑇𝐽 𝑀 → 𝑇 ′ 𝑀 is the isomorphism 𝜉(𝑣) = 𝑣 − 𝑖𝐽 𝑣.
The following statements are equivalent:
(a) 𝑔 is Kähler (i.e., 𝑑𝜔 = 0).
(b) For every 𝑝 ∈ 𝑀, there exists a real-valued function 𝑢 on a neighborhood
𝑉 of 𝑝 such that 𝜔|𝑉 = 𝑖𝜕𝜕𝑢.
(c) In every holomorphic coordinate chart, 𝜕𝑗 𝑔𝑙𝑘 = 𝜕𝑙 𝑔𝑗𝑘 for all 𝑗, 𝑘, 𝑙.
(d) For every 𝑎 ∈ 𝑀, there exists a holomorphic coordinate chart centered
at 𝑎 such that
(8.8) 𝑔𝑗𝑘 (𝑎) = 12 𝛿𝑗𝑘 and 𝑑(𝑔𝑗𝑘 )|𝑎 = 0.
(e) In every holomorphic coordinate chart, all Christoffel symbols of 𝑔 are
identically zero except those of types Γ𝑗𝑘
𝑙
and Γ𝑙𝑗𝑘 .
(f) ∇ restricts to the Chern connection on 𝑇 ′ 𝑀 with respect to the Hermitian
fiber metric ⟨⋅, ⋅⟩.
(g) For every real or complex vector field 𝑋, ∇𝑋 maps Γ(𝑇 ′ 𝑀) to itself.
(h) ∇𝐽 ≡ 0.
(i) ∇𝜔 ≡ 0.
(j) ∇𝜉 ≡ 0.
(k) ∇ is compatible with the holomorphic structure on 𝑇𝐽 𝑀.
(l) ∇(ℎ) = ∇.
(m) ∇(ℎ) is torsion-free.

Proof. Here is the plan:

(a)
KS ]e
+3 (b) +3 (c)
❇❇❇❇❇ ⑤⑤
❇❇❇ ⑤⑤⑤
z ⑤⑤⑤ 
(i)KS (d) (e)

(h)
KS \d
ks (g) ks (f)
❇❇❇❇❇
❇❇❇

(j) +3 (k) ks +3 (l) ks +3 (m).
Kähler Metrics 231

(a) ⇒ (b): This is the local 𝜕𝜕-lemma (Corollary 4.15).

(b) ⇒ (c): Suppose 𝜔 = 𝑖𝜕𝜕𝑢 on some open subset 𝑉 ⊆ 𝑀. In holomorphic
coordinates, 𝑖𝜕𝜕𝑢 = 𝑖𝜕𝑗 𝜕𝑘 𝑢 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑘 . Comparing this to (8.6), we find that 𝑔𝑗𝑘 =
𝜕𝑗 𝜕𝑘 𝑢, and therefore
𝜕𝑙 𝑔𝑗𝑘 = 𝜕𝑙 𝜕𝑗 𝜕𝑘 𝑢 = 𝜕𝑗 𝜕𝑙 𝜕𝑘 𝑢 = 𝜕𝑗 𝑔𝑙𝑘 ,
because the coordinate vector fields 𝜕𝑗 and 𝜕𝑙 commute.
(c) ⇒ (d): Given 𝑎 ∈ 𝑀, let (𝑧𝑗 ) be any holomorphic coordinates centered at
𝑎. Using these coordinates, we may as well consider 𝑔 as a Hermitian metric on
an open subset of ℂ𝑛 with 𝑎 = 0. Using the Gram-Schmidt algorithm, we can find
a preliminary complex-linear change of coordinates that achieves 𝑔𝑗𝑘 (0) = 12 𝛿𝑗𝑘 .
To eliminate the first derivatives of 𝑔 at 0, we will make a quadratic change of
coordinates by setting 𝑧 = 𝑓 (𝑤), where
𝑓 𝑗 (𝑤) = 𝑤𝑗 + 𝐴𝑚𝑙 𝑤𝑚 𝑤𝑙 ,
𝑗

for some constants 𝐴𝑚𝑙 to be determined, but assumed to be symmetric in 𝑚 and 𝑙.

𝑗

Note that this implies

𝜕𝑓 𝑗 𝜕2𝑓 𝑗
(0) = 𝛿𝑚 (0) = 2𝐴𝑚𝑙 .
𝑗 𝑗
𝜕𝑤𝑚 𝜕𝑤𝑚 𝑤𝑙
and

To compute the effect this has on 𝑔, we express 𝑔 in 𝑤-coordinates as 𝑔 =

𝑔 𝑙𝑚
̃ 𝑑𝑤𝑙 𝑑𝑤𝑚 , where
𝜕𝑓 𝑗 𝜕𝑓 𝑘
̃ = (𝑔𝑗𝑘 ∘ 𝑓 ) 𝑙
𝑔 𝑙𝑚 .
𝜕𝑤 𝜕𝑤𝑚
(Here 𝑤𝑚 and 𝑓 𝑘 represents the conjugates of 𝑤𝑚 and 𝑓 𝑘 , respectively.) Because
the holomorphic Jacobian of 𝑓 at the origin is the identity matrix, this change of
̃ (0) = 21 𝛿𝑙𝑚 .
coordinates maintains the condition 𝑔 𝑙𝑚

𝜕𝑔𝑗𝑘
By the chain rule (Prop. 1.47),
𝜕 𝑔 𝑙𝑚
̃ 𝜕𝑓 𝑝 𝜕𝑓 𝑗 𝜕𝑓 𝑘 𝜕 2 𝑓 𝑗 𝜕𝑓 𝑘
= ( 𝜕𝑧𝑝 ∘ 𝑓 ) 𝜕𝑤𝑞 𝜕𝑤𝑙 𝜕𝑤𝑚 + ( 𝑔 ∘ 𝑓 ) .
𝜕𝑤𝑞 𝑗𝑘
𝜕𝑤𝑞 𝜕𝑤𝑙 𝜕𝑤𝑚

𝑓 is holomorphic.) When we evaluate this at 0, we find

(The other terms that would ordinarily result from the chain rule are zero because

𝜕 𝑔 𝑙𝑚
̃ 𝜕𝑔𝑗𝑘
(0) = (0) 𝛿𝑞 𝛿𝑙 𝛿𝑚𝑘 + 𝑔𝑗𝑘 (0)2𝐴𝑞𝑙 𝛿𝑚𝑘
𝑝 𝑗 𝑗
𝜕𝑤 𝑞 𝜕𝑧𝑝
𝜕𝑔
= 𝑙𝑚 (0) + 2𝑔𝑗𝑚 (0)𝐴𝑞𝑙 .
𝑗
(8.9)

𝜕𝑧𝑞

1 𝜕𝑔
𝐴𝑞𝑙 = − 𝑔 𝑗𝑘 (0) 𝑙𝑘 (0),
Let us choose
𝑗
2 𝜕𝑧𝑞
where 𝑔 𝑗𝑘 = 𝑔 𝑘𝑗 is the matrix satisfying 𝑔𝑚𝑘 𝑔 𝑘𝑗 = 𝛿𝑚 . Note that hypothesis (c)
𝑗

guarantees that 𝐴𝑞𝑙 is symmetric in 𝑞 and 𝑙. It then follows easily from (8.9) (which
𝑗
232 8. Hermitian and Kähler Manifolds

implicitly used the symmetry of 𝐴𝑞𝑙 ) that each first derivative 𝜕 𝑔 𝑙𝑚

̃ /𝜕𝑤𝑞 vanishes
𝑗

at 0, and by conjugation so does 𝜕 𝑔 𝑙𝑚

̃ /𝜕𝑤𝑞 = 𝜕 𝑔 𝑚𝑙
̃ /𝜕𝑤𝑞 .
(d) ⇒ (a): Since 𝑑𝜔 is defined independently of coordinates, for each 𝑝 ∈ 𝑀
we can choose a holomorphic coordinate chart in which 𝑑(𝑔𝑗𝑘 ) = 0 at 𝑝, and then

𝑑𝜔𝑝 = 𝑑 (𝑖𝑔𝑗𝑘 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑘 )|𝑝 = 𝑖 𝑑𝑔𝑗𝑘 ∧ 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑘 |𝑝 = 0.

(c) ⇒ (e): Assuming (c), choose holomorphic coordinates (𝑧𝑗 ), and write the
Christoffel symbols of 𝑔 as
Γ𝑏𝑐
𝑎
= 𝑔 𝑎𝑑 (𝜕𝑏 𝑔𝑐𝑑 + 𝜕𝑐 𝑔𝑏𝑑 − 𝜕𝑑 𝑔𝑏𝑐 ),

where each of the indices 𝑎, 𝑏, 𝑐, 𝑑 ranges through {1, … , 𝑛, 1, … , 𝑛}. First note
that the matrices 𝑔 and 𝑔 −1 have the block form
0 𝑔𝑗𝑘 0 𝑔 𝑘𝑙
𝑔= , 𝑔 −1 = ,
(𝑔𝑗𝑘 0 ) (𝑔 𝑘𝑙 0 )

where (𝑔 𝑘𝑙 ) is the inverse of the matrix (𝑔𝑗𝑘 ) as above. We compute

1
Γ𝑗𝑘
𝑚
= 𝑔 𝑙𝑚 (𝜕𝑗 𝑔𝑘𝑙 + 𝜕𝑙 𝑔𝑘𝑗 − 𝜕𝑘 𝑔𝑗𝑙 ) = 0,
2
1
Γ𝑗𝑘
𝑚
= 𝑔 𝑙𝑚 (𝜕𝑗 𝑔𝑘𝑙 + 𝜕𝑘 𝑔𝑗𝑙 − 𝜕𝑙 𝑔𝑗𝑘 ) = 0,
2
where we have used the facts that 𝑔𝑗𝑘 = 0 since 𝑔 annihilates pairs sections of 𝑇 ′ 𝑀,
and 𝜕𝑙 𝑔𝑗𝑘 = 𝜕𝑗 𝑔𝑙𝑘 = 𝜕𝑗 𝑔𝑘𝑙 since we are assuming hypothesis (c). It then follows by
conjugation and symmetry that all other Christoffel symbols involving both barred

(e) ⇒ (f): First we need to check that ∇ maps Γ(𝑇 ′ 𝑀) to itself. This is just a
and unbarred indices are also zero.

computation in local holomorphic coordinates:

∇𝜕𝑘 𝜕𝑗 = Γ𝑘𝑗
𝑙
𝜕𝑙 + Γ𝑘𝑗
𝑙✚
𝜕𝑙 ,
✚

∇𝜕𝑘 𝜕𝑗 = Γ𝑘𝑗
✚𝜕𝑙 + Γ𝑙𝑘𝑗
𝑙 ✚
𝜕𝑙 .
✚

It is compatible with the Hermitian fiber metric ⟨⋅, ⋅⟩ on 𝑇 ′ 𝑀 because it is compat-

✚✚
✚ ✚

ible with 𝑔. To check that it is compatible with the holomorphic structure on 𝑇 ′ 𝑀,

suppose 𝑊 = 𝑊 𝑗 𝜕𝑗 is a holomorphic section of 𝑇 ′ 𝑀. Then for each 𝑘, we have

∇𝜕𝑘 𝑊 = (𝜕𝑘 𝑊 𝑗 )𝜕𝑗 + 𝑊 𝑗 Γ𝑙𝑘𝑗 𝜕𝑙 = 0 + 0.

(f) ⇒ (g): This is immediate.

(g) ⇒ (h): Suppose ∇ maps Γ(𝑇 ′ 𝑀) to itself. The covariant derivative of the
endomorphism field 𝐽 is defined by (∇𝑋 𝐽 )𝑌 = ∇𝑋 (𝐽 𝑌 ) − 𝐽 (∇𝑋 𝑌 ). If 𝑌 is a
section of 𝑇 ′ 𝑀, then so is ∇𝑋 𝑌 , and the formula for ∇𝑋 𝐽 (or rather its extension
Kähler Metrics 233

by complex linearity to complex vector fields) reduces to ∇𝑋 (𝐽 𝑌 ) − 𝐽 (∇𝑋 𝑌 ) =

∇𝑋 (𝑖𝑌 ) − 𝑖(∇𝑋 𝑌 ) = 0. Then by conjugation, the same formula holds whenever 𝑌
is a section of 𝑇 ″ 𝑀. Since 𝑇ℂ 𝑀 = 𝑇 ′ 𝑀 ⊕ 𝑇 ″ 𝑀, this proves that ∇𝐽 ≡ 0.
(h) ⇒ (i): The assumption ∇𝐽 ≡ 0 implies that ∇𝑉 (𝐽 𝑊 ) = 𝐽 ∇𝑉 𝑊 for all
complex vector fields 𝑉 and 𝑊 . Using the fact that the Levi-Civita connection is
compatible with 𝑔, we compute

(∇𝑉 𝜔)(𝑋, 𝑌 ) = 𝑉 (𝜔(𝑋, 𝑌 )) − 𝜔(∇𝑉 𝑋, 𝑌 ) − 𝜔(𝑋, ∇𝑉 𝑌 )

= 𝑉 (𝑔(𝐽 𝑋, 𝑌 )) − 𝑔 (𝐽 ∇𝑉 𝑋, 𝑌 ) − 𝑔 (𝐽 𝑋, ∇𝑉 𝑌 )
= 𝑔 (∇𝑉 𝐽 𝑋, 𝑌 )) + 𝑔 (𝐽 𝑋, ∇𝑉 𝑌 ))
− 𝑔 (𝐽 ∇𝑉 𝑋, 𝑌 ) − 𝑔 (𝐽 𝑋, ∇𝑉 𝑌 )
= 0.

(i) ⇒ (a): Let 𝑝 ∈ 𝑀 be arbitrary. As we did earlier in the proof, to prove

that 𝑑𝜔𝑝 = 0, we may use any convenient coordinate chart. If we use Riemannian
normal coordinates centered at 𝑝, then all of the Christoffel symbols vanish at 𝑝,
and the equation ∇𝜔 = 0 implies that all first partial derivatives of the components
of 𝜔 vanish at 𝑝. It follows easily that 𝑑𝜔|𝑝 = 0.
(h) ⇔ (j): With 𝑋 and 𝑌 arbitrary real vector fields, we compute

(∇𝑋 𝜉)𝑌 = ∇𝑋 (𝜉𝑌 ) − 𝜉(∇𝑋 𝑌 ) = ∇𝑋 (𝑌 − 𝑖𝐽 𝑌 ) − (∇𝑋 𝑌 − 𝑖𝐽 ∇𝑋 𝑌 )

= −𝑖(∇𝑋 (𝐽 𝑌 ) − 𝐽 ∇𝑋 𝑌 ) = −𝑖(∇𝑋 𝐽 )𝑌 .

Thus ∇𝑋 𝐽 = 0 for all 𝑋 if and only if ∇𝑋 𝜉 = 0 for all 𝑋.

(j) ⇒ (k): Note that the previous steps showed that (j) ⇒ (h) ⇒ ⋯ ⇒ (f), so
assuming ∇𝜉 ≡ 0 implies that ∇ is compatible with the holomorphic structure on
𝑇 ′ 𝑀. The way we have defined the holomorphic structure on 𝑇𝐽 𝑀 is via 𝜉, so a
section 𝑉 of 𝑇𝐽 𝑀 is holomorphic if and only if 𝜉(𝑉 ) is a holomorphic section of
𝑇 ′ 𝑀. Thus if 𝑉 is a holomorphic section of 𝑇𝐽 𝑀 and 𝑍 is a smooth section of
𝑇 ″ 𝑀, we have
𝜉(∇𝑍 𝑉 ) = ∇𝑍 (𝜉(𝑉 )) − (∇𝑍 𝜉)𝑉 = 0 − 0.
Since 𝜉 is injective, this proves that ∇𝑍 𝑉 = 0.
(k) ⇒ (h): Suppose ∇ is compatible with the holomorphic structure on 𝑇𝐽 𝑀.
Let 𝑧𝑗 = 𝑥𝑗 + 𝑖𝑦𝑗 be holomorphic local coordinates on 𝑀, and define smooth
sections (𝑋1 , … , 𝑋2𝑛 ) of 𝑇 𝑀 by
𝜕
⎧ 𝑗 , 𝑗 ≤ 𝑛,
⎪ 𝜕𝑥
𝑋𝑗 = ⎨
𝜕
⎪ , 𝑗 > 𝑛.
⎩ 𝜕𝑦𝑗
234 8. Hermitian and Kähler Manifolds

These form a commuting local frame for 𝑇 𝑀 (as a real vector bundle). A compu-

𝜕
tation shows that
⎧2 𝑗 , 𝑗 ≤ 𝑛,
⎪ 𝜕𝑧
𝜉(𝑋𝑗 ) = ⎨
⎪2𝑖 𝜕 , 𝑗 > 𝑛,
⎩ 𝜕𝑧𝑗
so by the way we have defined the holomorphic structure on 𝑇𝐽 𝑀, these are all
holomorphic sections of 𝑇𝐽 𝑀. Now for each 𝑗, note that 𝑋𝑗 + 𝑖𝐽 𝑋𝑗 is a section of
𝑇 ″ 𝑀, so the assumption that ∇ is compatible with the holomorphic structure on
𝑇𝐽 𝑀 implies
0 = ∇𝑋𝑗 +𝑖𝐽 𝑋𝑗 𝑋𝑘 = ∇𝑋𝑗 𝑋𝑘 + 𝐽 ∇𝐽 𝑋𝑗 𝑋𝑘 = ∇𝑋𝑘 𝑋𝑗 + 𝐽 ∇𝑋𝑘 (𝐽 𝑋𝑗 ),
where in the second equality we used the fact that multiplication by 𝑖 in 𝑇𝐽 𝑀 is
accomplished by applying 𝐽 ; and in the third we used the fact that ∇𝑉 𝑊 = ∇𝑊 𝑉
when 𝑉 and 𝑊 commute because ∇ is torsion-free. It follows that
(∇𝑋𝑘 𝐽 )𝑋𝑗 = ∇𝑋𝑘 (𝐽 𝑋𝑗 ) − 𝐽 ∇𝑋𝑘 𝑋𝑗 = −𝐽 (∇𝑋𝑘 𝑋𝑗 + 𝐽 ∇𝑋𝑘 (𝐽 𝑋𝑗 )) = 0,
which shows that 𝐽 is parallel with respect to ∇.
(k) ⇔ (l): The previous steps showed that (k) ⇒ (h) ⇒ (i), so assumption (k)
implies ∇𝜔 = 0. Therefore, ∇ℎ = ∇(𝑔 − 𝑖𝜔) = 0, so ∇ is also compatible with
the Hermitian fiber metric on 𝑇𝐽 𝑀. By uniqueness of the Chern connection, this
implies ∇ = ∇(ℎ) . Conversely, if ∇ = ∇(ℎ) , then ∇ is compatible with the holo-
morphic structure on 𝑇𝐽 𝑀 by definition of the Chern connection.
(l) ⇔ (m): If ∇(ℎ) = ∇, then ∇(ℎ) is torsion-free because ∇ is. Conversely,
suppose ∇(ℎ) is torsion-free. By definition of the Chern connection, for any real
vector field 𝑋, we have 0 = ∇𝑋 ℎ = ∇𝑋 (𝑔 − 𝑖𝜔). The real part of this equation
(ℎ) (ℎ)

is ∇𝑋 𝑔 = 0. Thus ∇(ℎ) is torsion-free and compatible with the metric 𝑔, so it is

(ℎ)

equal to the Levi-Civita connection of 𝑔. □

If 𝑔 is a Kähler metric with Kähler form 𝜔, a real-valued function 𝑢 such that

𝜔 = 𝑖𝜕𝜕𝑢 is called a Kähler potential for 𝒈. Part (b) of the preceding theorem shows
that every Kähler metric admits a Kähler potential in a neighborhood of each point;

never a global Kähler potential because 𝜔 is never exact.

but as we will see in Theorem 8.18 below, on a compact Kähler manifold there is

Examples of Kähler Metrics

There are plenty of examples of Kähler metrics.
Example 8.11 (The Standard Metric on ℂ𝒏 ). Let 𝑔E be the standard Hermitian
metric on ℂ𝑛 (see Example 8.6). Its fundamental 2-form is given by (8.2), which is
closed. Thus 𝑔E is Kähler. //
Examples of Kähler Metrics 235

the form ℂ𝑛 /Λ, where Λ ⊆ ℂ𝑛 is a lattice (see Example 1.18). Because the action
Example 8.12 (Complex Tori). Recall that a complex torus is a quotient space of

of Λ on ℂ𝑛 preserves the holomorphic structure, the Euclidean metric 𝑔E , and the

standard Kähler form 𝜔E = 𝑔E (⋅, 𝐽 ⋅), it follows that 𝑔E descends to a Kähler metric
on ℂ𝑛 /Λ. Thus every complex torus has a Kähler metric. //

manifold) has Hermitian metrics. Because every 2-form on a real 2-manifold is

Example 8.13 (Riemann Surfaces). Every Riemann surface (like every complex

closed for dimensional reasons, every Hermitian metric on a Riemann surface is

Kähler. //

Example 8.14 (The Fubini–Study Metric on ℂℙ𝒏 ). Recall the Hermitian fiber
metric ℎ we defined on the hyperplane bundle 𝐻 → ℂℙ𝑛 (Example 7.26). The
curvature of its Chern connection is the global 2-form Θ𝐻 whose expression in
each set 𝑈𝛼 where 𝑤𝛼 ≠ 0 is
|𝑤𝛼 |2
Θ𝐻 |𝑈𝛼 = 𝜕𝜕 log .
|𝑤|2
(8.10)

In affine coordinates (𝑧1 , … , 𝑧𝑛 ) ↔ [𝑧1 , … , 1, … , 𝑧𝑛 ] on 𝑈𝛼 , Example 7.26 showed

∑𝑗 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑗 ∑𝑗,𝑘 𝑧𝑗 𝑧𝑘 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑘

that

Θ𝐻 = − .
1 + |𝑧|2 (1 + |𝑧|2 )2
Let 𝜔FS be the following closed real (1, 1)-form:
𝑖
𝜔FS = Θ𝐻 .
2
(8.11)

We will show that 𝜔FS is positive, and therefore is a Kähler form. The correspond-
ing Kähler metric 𝑔FS = 𝜔FS (⋅, 𝐽 ⋅) is called the Fubini–Study metric, after the
mathematicians who first described it in the early twentieth century, Guido Fubini

each affine coordinate chart, it is given by the formula 𝑔FS = 2𝑔𝑗𝑘 𝑑𝑧𝑗 𝑑𝑧𝑘 , where
and Eduard Study [Fub04, Stu05]. It follows from the computation above that in

1 𝛿𝑗𝑘 𝑧𝑗 𝑧𝑘
𝑔𝑗𝑘 = − .
2 ( 1 + |𝑧|2 (1 + |𝑧|2 )2 )
(8.12)

(Some authors use a different normalization. Our definition of 𝑔FS is motivated

Problem 8-2. Others define the Kähler form to be equal to the Chern form of 𝐻,
primarily by the relationship with the standard metric on the sphere, described in

which is 1/𝜋 times our 𝜔FS ; the motivation for that choice is described in the proof

To see that 𝜔FS is a positive (1, 1)-form, or equivalently that 𝑔FS is positive
of Theorem 10.13.)

definite, let 𝑋 be a nonvanishing smooth real vector field on an open subset of one
of the affine charts 𝑈𝛼 , written as 𝑋 = 𝑍 + 𝑍 with 𝑍 a local section of 𝑇 ′ ℂℙ𝑛 ,
236 8. Hermitian and Kähler Manifolds

and compute

𝜔FS (𝑋, 𝐽 𝑋) = 𝜔FS (𝑍 + 𝑍, 𝑖𝑍 − 𝑖𝑍) = −2𝑖𝜔FS (𝑍, 𝑍) = Θ𝐻 (𝑍, 𝑍)

(1 + |𝑧|2 ) ∑𝑗 𝑍 𝑗 𝑍 𝑗 − ∑𝑗,𝑘 𝑧𝑗 𝑧𝑘 𝑍 𝑗 𝑍 𝑘
=
(1 + |𝑧|2 )2
(1 + |𝑧| )|𝑍| − |𝑍 ⋅ 𝑧|2
2 2
=
(1 + |𝑧|2 )2
(1 + |𝑧|2 )|𝑍|2 − |𝑍|2 |𝑧|2 |𝑍|2
≥ = > 0,
(1 + |𝑧|2 )2 (1 + |𝑧|2 )2

where we have written 𝑍 ⋅ 𝑧 = ∑𝑗 𝑍 𝑗 𝑧𝑗 and |𝑍|2 = 𝑍 ⋅ 𝑍, and the fourth line

follows from the Cauchy–Schwarz inequality. This shows that 𝜔FS is positive. //

mogeneous. Because the unitary group U(𝑛 + 1) ⊆ GL(𝑛 + 1, ℂ) acts transitively

One of the most important features of the Fubini–Study metric is that it is ho-

on 1-dimensional subspaces of ℂ𝑛+1 , it acts transitively on ℂℙ𝑛 by projective trans-

formations.

metric on ℂℙ𝑛 is invariant under the action of U(𝑛 + 1).

Proposition 8.15 (Homogeneity of the Fubini–Study Metric). The Fubini–Study

Proof. By (8.10) and (8.11), on 𝑈𝛼 ⊆ ℂℙ𝑛 we can write the Kähler form of 𝑔FS as
𝜔FS = 2𝑖 𝜕𝜕 log 𝑢𝛼 , where
|𝑤𝛼 |2
𝑢𝛼 ([𝑤]) = .
|𝑤|2
Suppose 𝐴 ∈ U(𝑛+1), and let 𝐴̃ ∶ ℂℙ𝑛 → ℂℙ𝑛 denote the corresponding projective
̃
transformation given by 𝐴([𝑤]) = [𝐴𝑤]. Given 𝑝 ∈ ℂℙ𝑛 , choose 𝛼, 𝛽 such that
̃ ∈ 𝑈𝛽 . Let 𝑓 = (𝐴𝑤)𝛽 /𝑤𝛼 , where (𝐴𝑤)𝛽 denotes the 𝛽 component
𝑝 ∈ 𝑈𝛼 and 𝐴(𝑝)
of 𝐴𝑤. Then 𝑓 is a nonvanishing holomorphic function on a neighborhood of 𝑝, so

̃ |(𝐴𝑤)𝛽 |2 |𝑓 |2 |𝑤𝛼 |2
𝑢𝛽 ∘ 𝐴([𝑤]) = = ,
|𝐴𝑤|2 |𝑤|2

where in the last equality we used the fact that |𝐴𝑤|2 = |𝑤|2 because 𝐴 is unitary.
Thus in a small enough neighborhood of 𝑝 where 𝑓 has a complex logarithm, we
have

𝑖 𝑖 |𝑤𝛼 |2
𝐴̃∗ 𝜔FS = 𝜕𝜕 log (𝑢𝛽 ∘ 𝐴̃) = 𝜕𝜕 (log 𝑓 + log 𝑓 + log = 𝜔FS ,
2 2 |𝑤|2 )
since 𝜕𝜕 log 𝑓 = 𝜕𝜕 log 𝑓 = 0. Therefore, 𝐴̃ preserves 𝜔FS , and since it also
preserves 𝐽 , it preserves 𝑔FS . □
Examples of Kähler Metrics 237

Here is one more class of examples of Kähler manifolds.

Example 8.16 (Kähler Submanifolds). Suppose 𝑀 is a Kähler manifold and
𝑁 ⊆ 𝑀 is a complex submanifold. Let 𝑔𝑀 , 𝜔𝑀 denote the Kähler metric and
Kähler form of 𝑀 and set 𝑔𝑁 = 𝜄∗ 𝑔𝑀 and 𝜔𝑁 = 𝜄∗ 𝜔𝑀 , where 𝜄 ∶ 𝑁 ↪ 𝑀 is the
inclusion map. Because 𝑑 commutes with pullbacks, 𝜔𝑁 is closed; and Proposition
4.10 shows that it is a (1, 1)-form. Since 𝑁 is a complex submanifold, the almost
complex structure map 𝐽 of 𝑀 maps 𝑇 𝑁 to itself. Thus for any vector field 𝑋 that
is tangent to 𝑁,
𝑔𝑁 (𝑋, 𝑋) = 𝑔𝑀 (𝑋, 𝑋) = 𝜔𝑀 (𝑋, 𝐽 𝑋) = 𝜔𝑁 (𝑋, 𝐽 𝑋),
so 𝑔𝑁 is a Kähler metric with Kähler form 𝜔𝑁 . Thus every complex submanifold
of a Kähler manifold is a Kähler manifold in a natural way.
In particular, this shows that all projective complex manifolds and all Stein
manifolds have Kähler metrics. //
On the other hand, there are complex manifolds that do not admit any Kähler
metrics. To exhibit some examples, we need the following lemma.
Lemma 8.17. Let (𝑀, 𝑔) be an 𝑛-dimensional Kähler manifold and 𝜔 its Kähler
form. The Riemannian volume form of 𝑔 is given explicitly by
1 𝑛 1
𝑑𝑉𝑔 = 𝜔 = 𝜔∧⋯∧𝜔
𝑛! 𝑛!
(𝑛-fold wedge product).

Proof. Given 𝑝 ∈ 𝑀, by Theorem 8.10(d) we can choose holomorphic coordinates

(𝑧𝑗 = 𝑥𝑗 + 𝑖𝑦𝑗 ) centered at 𝑝 such that 𝑔𝑗𝑘 (𝑝) = 21 𝛿𝑗𝑘 , which implies
𝑛 𝑛
𝑖
𝜔𝑝 = 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑗 = 𝑑𝑥𝑗 ∧ 𝑑𝑦𝑗 .
2∑𝑗=1
∑
𝑗=1

For 1 ≤ 𝑘 ≤ 𝑛, let 𝜔(𝑘) be the 2-form

𝑘
𝜔(𝑘) = 𝑑𝑥𝑗 ∧ 𝑑𝑦𝑗 .
∑
𝑗=1

𝜕 𝜕 𝜕 𝜕
Because the frame

( 𝜕𝑥1 , 𝜕𝑦1 , … , 𝜕𝑥𝑛 , 𝜕𝑦𝑛 )

is orthonormal at 𝑝 and positively oriented, it follows that
𝑑𝑉𝑔 |𝑝 = 𝑑𝑥1 ∧ 𝑑𝑦1 ∧ ⋯ ∧ 𝑑𝑥𝑛 ∧ 𝑑𝑦𝑛 |𝑝 .

We will prove by induction on 𝑘 that the following formula holds at 𝑝:

(8.13) (𝜔(𝑘) )𝑘 = 𝑘!𝑑𝑥1 ∧ 𝑑𝑦1 ∧ ⋯ ∧ 𝑑𝑥𝑘 ∧ 𝑑𝑦𝑘 .
238 8. Hermitian and Kähler Manifolds

For 𝑘 = 1 this is obvious, and for 𝑘 = 𝑛 it is the result we are trying to prove. Thus
suppose (8.13) holds for some 𝑘 < 𝑛, and write

𝜔(𝑘+1) = 𝛼 + 𝛽, where 𝛼 = 𝜔(𝑘) and 𝛽 = 𝑑𝑥𝑘+1 ∧ 𝑑𝑦𝑘+1 .

Note that 𝛼 and 𝛽 commute under wedge product because they are 2-forms. Also,
𝛽 ∧ 𝛽 = 0 because 𝑑𝑥𝑘+1 ∧ 𝑑𝑥𝑘+1 = 0, and the induction hypothesis shows that

𝛼 𝑘 = 𝑘! 𝑑𝑥1 ∧ 𝑑𝑦1 ∧ ⋯ ∧ 𝑑𝑥𝑘 ∧ 𝑑𝑦𝑘 .

It follows that 𝛼 𝑘+1 = 0 because it is a sum of terms each of which has a repeated
1-form.
Using the binomial theorem, we compute

(𝜔(𝑘+1) )𝑘+1 = (𝛼 + 𝛽)𝑘+1

(𝑘 + 1)𝑘 𝑘−1✘✘✘
=✟
𝛼 𝑘+1
✟ + (𝑘 + 1)𝛼 𝑘 ∧ 𝛽 + 𝛼 ✘ ∧ 𝛽2 + ⋯
2
✟
✘

+ (𝑘 + 1)✟
𝛼∧✟𝛽 𝑘 + ✟
𝛽 𝑘+1
✟ ✟
✟

= (𝑘 + 1)𝑘! 𝑑𝑥1 ∧ 𝑑𝑦1 ∧ ⋯ ∧ 𝑑𝑥𝑘 ∧ 𝑑𝑦𝑘 ∧ 𝑑𝑥𝑘+1 ∧ 𝑑𝑦𝑘+1 ,

thus completing the induction. □

Here is our first topological obstruction to the existence of Kähler metrics.

Theorem 8.18. If 𝑀 is a compact 𝑛-dimensional Kähler manifold with Kähler
form 𝜔, then 𝜔𝑘 = 𝜔 ∧ ⋯ ∧ 𝜔 represents a nonzero element of 𝐻dR 2𝑘
(𝑀; ℝ) for
𝑘 = 1, … , 𝑛. Thus a 2𝑛-dimensional compact manifold with 𝑏 (𝑀) = 0 for some
2𝑘

𝑘 ∈ {1, … , 𝑛} cannot admit Kähler metrics.

Proof. For each 𝑘 = 1, … , 𝑛, we have 𝑑(𝜔𝑘 ) = 𝑘 𝑑𝜔∧𝜔𝑘−1 = 0, so 𝜔𝑘 represents a

cohomology class in 𝐻dR2𝑘
(𝑀; ℝ). From the preceding lemma, we see that ∫𝑀 𝜔𝑛 =
𝑛! ∫𝑀 𝑑𝑉𝑔 > 0, so 𝜔𝑛 is not exact. Suppose for the sake of contradiction that 𝜔𝑘 is
exact for some 1 ≤ 𝑘 < 𝑛, so there is some 𝜂 such that 𝜔𝑘 = 𝑑𝜂. Then

𝜔𝑛 = 𝜔𝑘 ∧ 𝜔𝑛−𝑘 = 𝑑𝜂 ∧ 𝜔𝑛−𝑘 = 𝑑(𝜂 ∧ 𝜔𝑛−𝑘 ),

showing that 𝜔𝑛 is exact, a contradiction. Thus 𝜔𝑘 represents a nonzero element of

𝐻dR
2𝑘
(𝑀; ℝ) for each 𝑘 = 1, … , 𝑛. □

Example 8.19 (The 6-sphere). Problem 1-13 showed that there is a nonintegrable
almost complex structure on 𝕊6 , and noted that it is still not known whether there

structure, because 𝐻dR

2
(𝕊6 ; ℝ) = 𝐻dR
4
(𝕊6 ; ℝ) = 0.
is an integrable one. One thing we do know, however, is that there is no Kähler
//
Curvature of Kähler Metrics 239

plex 𝑛-manifolds diffeomorphic to 𝕊2𝑛−1 × 𝕊1 . It follows from the Künneth formula

Example 8.20 (Hopf Manifolds). The Hopf manifolds (Example 1.19) are com-

of algebraic topology [Hat02, Thm. 3.15] together with the de Rham theorem that

𝐻dR
2
(𝕊2𝑛−1 × 𝕊1 ; ℝ) ≅ (𝐻dR
2
(𝕊2𝑛−1 ; ℝ) ⊗ 𝐻dR
0
(𝕊1 ; ℝ))
⊕ (𝐻dR
1
(𝕊2𝑛−1 ; ℝ) ⊗ 𝐻dR
1
(𝕊1 ; ℝ)) ⊕ (𝐻dR
0
(𝕊2𝑛−1 ; ℝ) ⊗ 𝐻dR
2
(𝕊1 ; ℝ)),

which is zero when 𝑛 > 1 because all three terms in the direct sum are zero. Thus by
Theorem 8.18, Hopf manifolds of complex dimension greater than 1 do not admit
Kähler metrics. //

Curvature of Kähler Metrics

Suppose (𝑀, 𝑔) is a Kähler manifold, and let 𝑅𝑚 denote its Riemann curvature
tensor, a covariant 4-tensor field defined by

𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍) = 𝑔(𝑅(𝑊 , 𝑋)𝑌 , 𝑍),

where 𝑅 ∶ Γ(𝑇 𝑀) × Γ(𝑇 𝑀) × Γ(𝑇 𝑀) → Γ(𝑇 𝑀) is the curvature endomorphism

field defined by

𝑅(𝑊 , 𝑋)𝑌 = ∇𝑊 ∇𝑋 𝑌 − ∇𝑋 ∇𝑊 𝑌 − ∇[𝑊 ,𝑋] 𝑌 .

Initially defined only for real vector fields, 𝑅 and 𝑅𝑚 can be extended to act on

over 𝐶 ∞ (𝑀; ℂ), and satisfies the standard Riemannian symmetries for all complex
complex vector fields by the same formulas. The Riemann curvature is multilinear

vector fields 𝑊 , 𝑋, 𝑌 , 𝑍:

𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍) = −𝑅𝑚(𝑋, 𝑊 , 𝑌 , 𝑍),

𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍) = −𝑅𝑚(𝑊 , 𝑋, 𝑍, 𝑌 ),
(8.14)

𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍) = 𝑅𝑚(𝑌 , 𝑍, 𝑊 , 𝑋),

(8.15)

𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍) + 𝑅𝑚(𝑋, 𝑌 , 𝑊 , 𝑍) + 𝑅𝑚(𝑌 , 𝑊 , 𝑋, 𝑍) = 0.

(8.16)
(8.17)

(Equation (8.17) is called the algebraic Bianchi identity.) These are proved for

vector fields, just note that we can choose a local frame for 𝑇ℂ 𝑀 consisting of real
real vector fields in [LeeRM, Prop. 7.12]. To see that they also hold for complex

by complex multilinearity. In addition, since the action of 𝑅𝑚 on complex vector

vector fields, so the identities for complex vector fields follow from the real ones

fields is obtained from its action on real ones by complex multilinearity, we have

(8.18) 𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍) = 𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍).

240 8. Hermitian and Kähler Manifolds

On a Kähler manifold, there are additional symmetries.

Theorem 8.21 (Kähler Curvature Symmetries). In addition to the standard sym-

for all 𝑊 , 𝑋, 𝑌 , 𝑍 ∈ Γ(𝑇 ′ 𝑀):

metries, the curvature tensor of a Kähler metric satisfies the following symmetries

𝑅𝑚(𝑊 , 𝑋, ⋅, ⋅) = 𝑅𝑚(⋅, ⋅, 𝑊 , 𝑋) = 0,
𝑅𝑚(𝑊 , 𝑋, ⋅, ⋅) = 𝑅𝑚(⋅, ⋅, 𝑊 , 𝑋) = 0
(8.19)
(8.20)
(8.21) 𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍) = 𝑅𝑚(𝑌 , 𝑋, 𝑊 , 𝑍),
(8.22) 𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍) = 𝑅𝑚(𝑊 , 𝑍, 𝑌 , 𝑋),

Proof. Suppose 𝑍, 𝑊 are sections of 𝑇 ′ 𝑀 and 𝑈 , 𝑉 are arbitrary complex vector

fields. Then 𝑅(𝑈 , 𝑉 )𝑍 = ∇𝑈 ∇𝑉 𝑍 − ∇𝑉 ∇𝑈 𝑍 − ∇[𝑈 ,𝑉 ] 𝑍 ∈ Γ(𝑇 ′ 𝑀) because the
Levi-Civita connection maps Γ(𝑇 ′ 𝑀) to itself (Thm. 8.10(g)). Hence
𝑅𝑚(𝑈 , 𝑉 , 𝑍, 𝑊 ) = 𝑔(𝑅(𝑈 , 𝑉 )𝑍, 𝑊 ) = 0,
because 𝑔 annihilates pairs of sections of 𝑇 ′ 𝑀. This proves (8.19), and then (8.20)

For sections 𝑊 , 𝑋, 𝑌 , 𝑍 of 𝑇 ′ 𝑀, the algebraic Bianchi identity (8.17) gives

follows by conjugation.

0 = 𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍) + 𝑅𝑚(𝑋, 𝑌 , 𝑊 , 𝑍) + ✭
𝑅𝑚(𝑌
✭✭,✭𝑊✭, ✭
𝑋, 𝑍)
✭✭

= 𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍) − 𝑅𝑚(𝑌 , 𝑋, 𝑊 , 𝑍),

which is (8.21), and then (8.22) follows from (8.16). □

► Exercise 8.22. Prove that the curvature tensor on a Kähler manifold satisfies
the following additional symmetries for all (real or complex) vector fields:
𝑅𝑚(𝑊 , 𝑋, 𝐽 𝑌 , 𝐽 𝑍) = 𝑅𝑚(𝑊 , 𝑋, 𝑌 , 𝑍) = 𝑅𝑚(𝐽 𝑊 , 𝐽 𝑋, 𝑌 , 𝑍).

coordinates. Using the facts that 𝑔𝑙𝑝 = 0 and 𝜕𝑗 𝑔𝑙𝑞 = 𝜕𝑙 𝑔𝑗𝑞 , we compute
Let us work out the formulas for the connection and curvature in holomorphic

Γ𝑗𝑙 = 21 𝑔 𝑝𝑞 (𝜕𝑗 𝑔𝑙𝑞 + 𝜕𝑙 𝑔𝑗𝑞 − ✟

𝜕𝑞 ✟
𝑔✟
𝑝
𝑙𝑝 )
= 𝑔 𝜕𝑗 𝑔𝑙𝑞 .
𝑝𝑞
(8.23)

(This can also be viewed as a reflection of the fact that ∇ restricts to the Chern
connection on 𝑇 ′ 𝑀 by virtue of Theorem 8.10(f), so its connection 1-forms are
determined by (7.17).) The only other nonzero Christoffel symbols are the ones
obtained from these by conjugation.
Using this formula, we can compute the components of the curvature endomor-
phism field and the Riemann curvature tensor.
𝑅(𝜕𝑗 , 𝜕𝑘 )𝜕𝑙 = ✘
∇𝜕𝑗
✘∇ 𝜕 − ∇𝜕𝑘 ∇𝜕𝑗 𝜕𝑙 − ✘
𝑘 𝑙
✘𝜕 ✘✘ ∇[𝜕
✘𝑗 ,𝜕✘ 𝜕
𝑘] 𝑙
✘
✘

= −𝜕𝑘 (Γ𝑗𝑙 )𝜕𝑝 − ✘ ∇𝜕𝑘 𝜕𝑝 ,

Γ𝑗𝑙✘
𝑝 𝑝 ✘✘
Curvature of Kähler Metrics 241

so

𝑅𝑗𝑘𝑙 𝑝 = −𝜕𝑘 (Γ𝑗𝑙 ) = −𝜕𝑘 (𝑔 𝑝𝑞 𝜕𝑗 𝑔𝑙𝑞 ),

𝑝
(8.24)
𝑅𝑗𝑘𝑙𝑚 = −𝑔𝑝𝑚 𝜕𝑘 (Γ𝑗𝑙 ) = −𝑔𝑝𝑚 𝜕𝑘 (𝑔 𝑝𝑞 𝜕𝑗 𝑔𝑙𝑞 ).
𝑝
(8.25)

We will use these formulas later in the chapter.

Holomorphic Sectional Curvature

to the sectional curvature. For each nonzero vector 𝑍 ∈ 𝑇 ′ 𝑀, we define the holo-
A Kähler metric allows us to define a new curvature invariant, closely related

morphic sectional curvature in the direction 𝑍 by

𝑅𝑚(𝑍, 𝑍, 𝑍, 𝑍 )
𝐻(𝑍) = .
|𝑍|4
Lemma 8.23. For every nonzero vector 𝑍 ∈ 𝑇 ′ 𝑀, 𝐻(𝑍) is the ordinary sectional
curvature of the plane spanned by {Re 𝑍, Im 𝑍}.

Proof. Given 𝑍 ∈ 𝑇 ′ 𝑀, let us write 𝑋 = Re 𝑍 and 𝑌 = Im 𝑍. The sectional

curvature of the plane spanned by {𝑋, 𝑌 } is
𝑅𝑚(𝑋, 𝑌 , 𝑌 , 𝑋)
sec(𝑋, 𝑌 ) = .
|𝑋|2 |𝑌 |2 − ⟨𝑋, 𝑌 ⟩2

using 𝑋 = 21 (𝑍 + 𝑍) and 𝑌 = 2𝑖1 (𝑍 − 𝑍), and interpreting the inner product in the
(See [LeeRM, Prop. 8.29].) We will prove the lemma by expanding this formula

denominator as the Hermitian fiber metric on 𝑇ℂ 𝑀. The Kähler symmetries of the

curvature tensor allow us to simplify the numerator:

𝑅𝑚(𝑋, 𝑌 , 𝑌 , 𝑋) = − 18 (𝑅𝑚(𝑍 + 𝑍, 𝑍 − 𝑍, 𝑍 − 𝑍, 𝑍 + 𝑍)
= − 18 (𝑅𝑚(𝑍, −𝑍, 𝑍, 𝑍) + 𝑅𝑚(𝑍, −𝑍, −𝑍, 𝑍)
+ 𝑅𝑚(𝑍, 𝑍, 𝑍, 𝑍) + 𝑅𝑚(𝑍, 𝑍, −𝑍, 𝑍))
= 12 𝑅𝑚(𝑍, 𝑍, 𝑍, 𝑍).

On the other hand, for the denominator, we use the facts that 𝑇 ′ 𝑀 and 𝑇 ″ 𝑀 are
orthogonal and |𝑍|2 = |𝑍|2 to compute

|𝑋|2 |𝑌 |2 − ⟨𝑋, 𝑌 ⟩2 = 81 (|𝑍 + 𝑍|2 |𝑍 − 𝑍|2 − ⟨𝑍 + 𝑍, 𝑍 − 𝑍⟩2 )

= |𝑍|2 + |𝑍|2 )(|𝑍|2 + |𝑍|2 ) − (|𝑍|2 − |𝑍|2 ) )

1 2
8 ((

= 1
2
|𝑍|4 . □
242 8. Hermitian and Kähler Manifolds

Although this lemma suggests that the holomorphic sectional curvatures con-
tain only part of the information encoded by the curvature tensor (namely, sectional

tors, or equivalently by pairs of real vectors of the form {𝑋, 𝐽 𝑋}), it turns out that
curvatures of planes spanned by the real and imaginary parts of holomorphic vec-

they actually determine the full curvature tensor thanks to the Kähler symmetries.

Suppose 𝑅1 and 𝑅2 are covariant 4-tensor fields on a complex manifold 𝑀 that

Lemma 8.24 (Holomorphic Sectional Curvature Determines the Curvature).

satisfy the Riemann curvature symmetries (8.14)–(8.17) as well as the Kähler sym-

𝑍 ∈ 𝑇 ′ 𝑀,
metries (8.19)–(8.22). If the following equality holds for every nonzero vector

𝑅1 (𝑍, 𝑍, 𝑍, 𝑍) 𝑅2 (𝑍, 𝑍, 𝑍, 𝑍)
= ,
|𝑍|4 |𝑍|4
then 𝑅1 = 𝑅2 .

Proof. Define 𝐺 = 𝑅1 − 𝑅2 , so that 𝐺 has the same symmetries as 𝑅1 and 𝑅2 and

satisfies 𝐺(𝑍, 𝑍, 𝑍, 𝑍) = 0 for all 𝑍 ∈ 𝑇 ′ 𝑀. Given a point 𝑝 ∈ 𝑀 and vectors
𝑍, 𝑊 ∈ 𝑇𝑝′ 𝑀, consider the following smooth function of a complex variable 𝑧:

𝑢(𝑧) = 𝐺(𝑊 + 𝑧𝑍, 𝑊 + 𝑧𝑍, 𝑊 + 𝑧𝑍, 𝑊 + 𝑧𝑍 ).

Our hypothesis guarantees that 𝑢 is identically zero, so using the Kähler symmetries

𝜕2𝑢
we conclude
0= (0) = 4𝐺(𝑊 , 𝑊 , 𝑍, 𝑍 ).
𝜕𝑧𝜕𝑧
Now let 𝑈 , 𝑉 be two more elements of 𝑇𝑝′ 𝑀 and consider the smooth function
𝑣(𝑤, 𝑧) defined by
𝑣(𝑤, 𝑧) = 𝐺(𝑊 + 𝑤𝑈 , 𝑊 + 𝑤𝑈 , 𝑍 + 𝑧𝑉 , 𝑍 + 𝑧𝑉 ).
The first part of the proof showed that 𝑣 is identically zero, so
𝜕2𝑣
0= (0, 0) = 𝐺(𝑈 , 𝑊 , 𝑉 , 𝑍 ).
𝜕𝑤𝜕𝑧
It then follows from the Riemann and Kähler symmetries that 𝐺 = 0. □

if there is a constant 𝑐 such that 𝐻(𝑍) = 𝑐 for every 𝑝 ∈ 𝑀 and every nonzero
A Kähler manifold is said to have constant holomorphic sectional curvature

𝑍 ∈ 𝑇𝑝′ 𝑀.
Lemma 8.25. A Kähler metric 𝑔 has constant holomorphic sectional curvature 𝑐
if and only if in each holomorphic coordinate chart, the coefficients of the Riemann
curvature tensor satisfy
(8.26) 𝑅𝑗𝑘𝑙𝑚 = 12 𝑐(𝑔𝑗𝑘 𝑔𝑙𝑚 + 𝑔𝑙𝑘 𝑔𝑗𝑚 ).
Curvature of Kähler Metrics 243

then 𝑅𝑚(𝑍, 𝑍, 𝑍, 𝑍) = 𝑐|𝑍|4 for all 𝑍 ∈ 𝑇 ′ 𝑀, and therefore 𝑔 has constant

Proof. A simple computation shows that if the curvature tensor satisfies (8.26),

holomorphic sectional curvature 𝑐. Conversely, if 𝑔 has constant holomorphic sec-

tional curvature 𝑐, then the curvature tensor of 𝑔 and the one given by the right-hand
side of (8.26) both give the same holomorphic sectional curvatures, so they must
be equal by Lemma 8.24. □

Example 8.26 (Constant Holomorphic Sectional Curvature Manifolds).

(a) The standard Kähler metric on ℂ𝑛 (which is just the Euclidean metric) is

(b) We will show that the Fubini–Study metric on ℂℙ𝑛 has constant holo-
flat, and thus has constant holomorphic sectional curvature zero.

morphic sectional curvature equal to 4. To verify this, because the met-

point [1, 0, … , 0], which corresponds to the origin in affine coordinates

ric is homogeneous, it suffices to check it at one point. We choose the

(𝑧1 , … , 𝑧𝑛 ) ↔ [1, 𝑧1 , … , 𝑧𝑛 ]. It follows from (8.12) that

1 𝛿𝑙𝑚 𝑧𝑗 𝛿𝑗𝑚 𝑧 𝑙
2𝑧𝑙 𝑧𝑚 𝑧𝑗
𝜕𝑗 𝑔𝑙𝑚 = (− − + .
2 (1 + |𝑧|2 )2 (1 + |𝑧|2 )2 (1 + |𝑧|2 )3 )
Therefore, at the origin, 𝑔𝑙𝑚 = 21 𝛿𝑙𝑚 , 𝜕𝑗 𝑔𝑙𝑚 = 𝜕𝑘 𝑔𝑙𝑚 = 0, and

𝜕𝑘 𝜕𝑗 𝑔𝑙𝑚 = − 12 (𝛿𝑙𝑚 𝛿𝑗𝑘 + 𝛿𝑗𝑚 𝛿𝑙𝑘 ).

Thus it follows from the result of Problem 8-11 that
𝑅𝑗𝑘𝑙𝑚 = 21 (𝛿𝑙𝑚 𝛿𝑗𝑘 + 𝛿𝑗𝑚 𝛿𝑙𝑘 ) = 2(𝑔𝑗𝑘 𝑔𝑙𝑚 + 𝑔𝑙𝑘 𝑔𝑗𝑚 ),

origin are all equal to 4. The result then follows by homogeneity.

and Lemma 8.25 shows that the holomorphic sectional curvatures at the

(c) The complex hyperbolic metric is the Kähler metric 𝑔CH on 𝔹2𝑛 whose

𝑖
Kähler form is
𝜔CH = − 𝜕𝜕 log(1 − |𝑧|2 ).
2
Problem 8-5 asks you to show that 𝑔CH is homogeneous and geodesically
complete, and has constant holomorphic sectional curvature −4. //
Thus we have examples of Kähler manifolds in every dimension with constant
holomorphic sectional curvature that is positive (the Fubini–Study metric), zero (the
Euclidean metric), and negative (the complex hyperbolic metric). By multiplying
these metrics by constants, we can obtain a Kähler metric with holomorphic sec-
tional curvature equal to any real constant. The next theorem shows that, in a sense,
these are all the possibilities. It is based on the theory of analytic continuations of

the following: Suppose (𝑀, 𝑔) and (𝑀, ̂ 𝑔 )̂ are Riemannian manifolds of the same
Riemannian isometries described in [LeeRM, Chap. 12]. The basic definition is

dimension and 𝜑 ∶ 𝑈 → 𝑀 ̂ is a local isometry (that is, a smooth map such that
244 8. Hermitian and Kähler Manifolds

𝜑∗ 𝑔 ̂ = 𝑔) defined on a connected open subset 𝑈 ⊆ 𝑀. For any continuous path

𝛾 ∶ [0, 1] → 𝑀 such that 𝛾(0) ∈ 𝑈 , an analytic continuation of 𝝋 along 𝜸 is a
family of pairs {(𝑈𝑡 , 𝜑𝑡 ) ∶ 𝑡 ∈ [0, 1]}, where 𝑈𝑡 is a connected neighborhood of
𝛾(𝑡) and 𝜑𝑡 ∶ 𝑈𝑡 → 𝑀 ̂ is a local isometry, such that 𝜑0 = 𝜑 on 𝑈0 ∩ 𝑈 , and for
each 𝑡 ∈ [0, 1] there exists 𝛿 > 0 such that |𝑡 − 𝑡1 | < 𝛿 implies that 𝛾(𝑡1 ) ∈ 𝑈𝑡 and
that 𝜑𝑡 agrees with 𝜑𝑡1 on 𝑈𝑡 ∩ 𝑈𝑡1 .
If (𝑀, 𝑔) and (𝑀, ̂ 𝑔 )̂ are Riemannian manifolds, a smooth covering map
̂
𝜋 ∶ 𝑀 → 𝑀 is called a Riemannian covering if it is also a local isometry.
Theorem 8.27. Suppose (𝑀, 𝑔) is an 𝑛-dimensional Kähler manifold with constant
holomorphic sectional curvature 𝑐, and assume 𝑀 is connected and geodesically
complete. Then 𝑀 admits a holomorphic Riemannian covering by one of the fol-
lowing:
• ℂℙ𝑛 with a multiple of the Fubini–Study metric,
• ℂ𝑛 with the Euclidean metric, or
• 𝔹2𝑛 with a multiple of the complex hyperbolic metric.

Proof. First consider the case in which 𝑀 is simply connected. Given (𝑀, 𝑔)
satisfying the hypothesis, note that because the Riemann curvature tensor can be

∇𝑅𝑚 ≡ 0 since the metric is parallel. Let (𝑀, ̂ 𝑔)̂ be the model Kähler manifold
expressed purely in terms of the metric tensor by formula (8.25), it follows that

(ℂℙ , ℂ , or 𝔹 ) endowed with a Kähler metric 𝑔 ̂ with the same constant holo-
𝑛 𝑛 2𝑛

morphic sectional curvature as 𝑀. Then (𝑀, ̂ 𝑔)̂ also has a parallel curvature tensor

Given any points 𝑝 ∈ 𝑀 and 𝑝 ̂ ∈ 𝑀, ̂ let 𝐴 ∶ 𝑇𝑝 𝑀 → 𝑇𝑝 𝑀 ̂ be a complex-linear

by the same reasoning, and is geodesically complete and simply connected.
̂
̂
isometry. Because 𝑀 and 𝑀 both have the same constant holomorphic sectional
curvature, it follows from (8.25) that 𝐴∗ (𝑅𝑚𝑝 )̂ = 𝑅𝑚𝑝 . Then [LeeRM, Lemma
10.18] shows that there is a Riemannian isometry 𝜑 ∶ 𝑈 → 𝑈 ̂ from a neighborhood
𝑈 of 𝑝 to a neighborhood 𝑈 ̂ of 𝑝 ̂ that satisfies 𝜑(𝑝) = 𝑝 ̂ and 𝐷𝜑(𝑝) = 𝐴 (where
𝐷𝜑(𝑝) is the total derivative). (That lemma is stated for two points in the same

be the disjoint union of 𝑀 and 𝑀.) ̂ Corollary 12.3 of [LeeRM] shows that if 𝜑
Riemannian manifold, but it applies in the present case by taking the manifold to

can be analytically continued along every path starting at 𝑝, then there is a global
Riemannian isometry from 𝑀 to 𝑀 ̂ that agrees with 𝜑 on a neighborhood of 𝑝.

isometry between open subsets of 𝑀 ̂ is the restriction of a global isometry (Problem

To see that analytic continuation is always possible, we use the fact that every

8-6). Let 𝛾 ∶ [0, 1] → 𝑀 be a path starting at 𝑝. Each point in the image of 𝛾 has a
geodesically convex neighborhood that is isometric to an open subset of 𝑀. ̂ Cover
the image of 𝛾 with finitely many such neighborhoods 𝑈0 , … , 𝑈𝑘 , with 𝑝 ∈ 𝑈0 ⊆
𝑈 , 𝑈𝑖 ∩ 𝑈𝑖+𝑖 ≠ ∅ for each 𝑖, and 𝛾(1) ∈ 𝑈𝑘 . For each 𝑖, let 𝜑𝑖 ∶ 𝑈𝑖 → 𝑈 ̂𝑖 ⊆ 𝑀̂
be an isometry, chosen so that 𝜑0 = 𝜑|𝑈0 . For each 𝑖, the intersection 𝑈𝑖 ∩ 𝑈𝑖+1 is
Ricci and Scalar Curvatures 245

connected, and the composition 𝜑𝑖 ∘𝜑−1 𝑖+1 is an isometry between the connected open
̂ so by Problem 8-6 it is the restriction
sets 𝜑𝑖+1 (𝑈𝑖 ∩ 𝑈𝑖+1 ) and 𝜑𝑖 (𝑈𝑖 ∩ 𝑈𝑖+1 ) in 𝑀,
of a global isometry Ψ. By replacing 𝜑𝑖+1 by Ψ ∘ 𝜑𝑖+1 , we can ensure that 𝜑𝑖+1
agrees with 𝜑𝑖 on the overlap. Thus by induction, we can continue 𝜑 all the way to
𝛾(1).
Consequently, there exists a global Riemannian isometry Φ ∶ 𝑀 → 𝑀. ̂ To see
̂
that it is holomorphic, we argue as follows. Let 𝐽 and 𝐽 be the almost complex
structures of 𝑀 and 𝑀, ̂ respectively, and define Φ∗ 𝐽̂∶ 𝑇 𝑀 → 𝑇 𝑀 by

(Φ∗ 𝐽̂)(𝑋𝑞 ) = 𝐷Φ(𝑞)−1 𝐽̂(𝐷Φ(𝑞)𝑋𝑞 ).

Because Φ is an isometry, it pulls back the Levi-Civita connection of 𝑀 ̂ to that of
̂ ∗̂
𝑀, and because 𝐽 is parallel, it follows that Φ 𝐽 is also parallel. At the original
point 𝑝 ∈ 𝑀, we have 𝐷Φ(𝑝) = 𝐷𝜑(𝑝) = 𝐴, which was chosen to be complex-
linear; thus Φ∗ 𝐽̂ agrees with 𝐽 at 𝑝. Since both Φ∗ 𝐽̂ and 𝐽 are parallel and 𝑀 is
connected, it follows that Φ∗ 𝐽̂ ≡ 𝐽 , and therefore Φ is holomorphic. That com-
pletes the proof in case 𝑀 is simply connected.

space of 𝑀 with the pullback metric.

For the general case, just apply the above argument to the universal covering
□

Ricci and Scalar Curvatures

On any Riemannian manifold, the Ricci curvature is the covariant 2-tensor field

𝑅𝑐(𝑋, 𝑌 ) = tr (𝑍 ↦ 𝑅(𝑍, 𝑋)𝑌 ).

defined by

Because the trace of a linear map is well defined, independent of choice of basis,
this is a globally defined tensor field, and it follows from the symmetries of the

any local frame, it has components 𝑅𝑎𝑏 = 𝑅𝑐𝑎𝑏 𝑐 . In addition, the scalar curvature
Riemann curvature tensor that it is symmetric [LeeRM, Lemma 7.15]). In terms of

is the real-valued function 𝑆 defined by raising an index of 𝑅𝑐 and taking the trace;
in a local frame, it is 𝑆 = 𝑔 𝑎𝑏 𝑅𝑎𝑏 .
Lemma 8.28. On a Kähler manifold in holomorphic coordinates, the Ricci and
scalar curvatures have the coordinate expressions
(8.27) 𝑅𝑐 = 2𝑅𝑗𝑘 𝑑𝑧𝑗 𝑑𝑧𝑘 , 𝑆 = 2𝑔 𝑗𝑘 𝑅𝑗𝑘 ,
where the coefficients 𝑅𝑗𝑘 are given by any of the following expressions:
(8.28) 𝑅𝑗𝑘 = 𝑅𝑚𝑗𝑘 𝑚 = 𝑅𝑗𝑘𝑙 𝑙 = −𝜕𝑗 𝜕𝑘 log(det 𝑔).
(Here det 𝑔 denotes the determinant of the 𝑛 × 𝑛 matrix (𝑔𝑗𝑘 ) in coordinates.)

Proof. The formula for the scalar curvature is an easy consequence of the one for
the Ricci curvature, so we focus on the latter.
246 8. Hermitian and Kähler Manifolds

A priori, as a symmetric 2-tensor field, the Ricci curvature might have compo-
nents of the forms 𝑅𝑗𝑘 , 𝑅𝑗𝑘 = 𝑅𝑘𝑗 , and 𝑅𝑗𝑘 . However, on a Kähler manifold,
𝑅𝑗𝑘 = 𝑅𝑙𝑗𝑘 𝑙 + 𝑅𝑙𝑗𝑘 𝑙 = 𝑔 𝑙𝑚 𝑅𝑙𝑗𝑘𝑚 + 𝑔 𝑚𝑙 𝑅𝑙𝑗𝑘𝑚 = 0 + 0,
and by conjugation 𝑅𝑗𝑘 = 0 as well. Thus the only nontrivial components are those
of the form 𝑅𝑗𝑘 , so we can write 𝑅𝑐 as in (8.27) with 𝑅𝑗𝑘 = 𝑅𝑐(𝜕𝑗 , 𝜕𝑘 ).
For the components of the Ricci tensor, we use the Kähler curvature symmetries
to compute
𝑅𝑗𝑘 = 𝑅𝑚𝑗𝑘
✟✟ + 𝑅𝑚𝑗𝑘
𝑚 𝑚

= 𝑔 𝑙𝑚 𝑅𝑚𝑗𝑘𝑙 = 𝑔 𝑙𝑚 𝑅𝑗𝑚𝑙𝑘 = 𝑔 𝑙𝑚 𝑅𝑗𝑘𝑙𝑚

✟

= 𝑅𝑗𝑘𝑙 𝑙 .
This proves the first two equalities in (8.28).
The total derivative of the determinant function is given by
𝐷(det)𝑋 (𝐵) = (det 𝑋) tr (𝑋 −1 𝐵)
for 𝑋 ∈ GL(𝑛, ℂ) and any 𝑛 × 𝑛 matrix 𝐵. (See [LeeSM, Problem 7-4, p. 172],
where the proof is sketched for GL(𝑛, ℝ); the same argument works for complex
matrices.) Thus by the chain rule,
(8.29) 𝜕𝑗 (det 𝑔) = 𝐷(det)𝑔 (𝜕𝑗 𝑔) = (det 𝑔) tr (𝑔 −1 𝜕𝑗 𝑔 ) = (det 𝑔)𝑔 𝑙𝑞 𝜕𝑗 𝑔𝑙𝑞 .
Now, (8.23) shows that
(8.30) Γ𝑗𝑙𝑙 = 𝑔 𝑙𝑞 𝜕𝑗 𝑔𝑙𝑞 = 𝜕𝑗 log(det 𝑔).
On the other hand, (8.24) shows that
(8.31) 𝑅𝑗𝑘 = 𝑅𝑗𝑘𝑙 𝑙 = −𝜕𝑘 (Γ𝑗𝑙𝑙 ).

that 𝜕𝑗 and 𝜕𝑘 commute.

The third equality in (8.28) follows from (8.29) and (8.30), together with the fact
□

This leads to an important feature of the Ricci curvature.

Lemma 8.29. On a Kähler manifold, the Ricci tensor is invariant under 𝐽 : for all
complex vector fields 𝑋 and 𝑌 ,
𝑅𝑐(𝐽 𝑋, 𝐽 𝑌 ) = 𝑅𝑐(𝑋, 𝑌 ).

roles of (𝑋, 𝑌 ) played by (𝜕𝑗 , 𝜕𝑘 ), (𝜕𝑗 , 𝜕𝑘 ) and (𝜕𝑗 , 𝜕𝑘 ).

Proof. This follows straightforwardly in coordinates by applying (8.27) with the
□

As a consequence of this lemma, we can “twist” the Ricci tensor with 𝐽 to pro-
duce a 2-form, in the same way we did with the metric to produce the fundamental
2-form.
Ricci and Scalar Curvatures 247

Proposition 8.30 (The Ricci Form). Let (𝑀, 𝑔) be a Kähler manifold and 𝑅𝑐 its
Ricci curvature. Define a 2-tensor field 𝜌 by
𝜌(𝑋, 𝑌 ) = 𝑅𝑐(𝐽 𝑋, 𝑌 ).
Then 𝜌 is a closed (1, 1)-form, called the Ricci form of 𝒈.

Proof. To see that 𝜌 is antisymmetric, we use Lemma 8.29:

𝜌(𝑋, 𝑌 ) = 𝑅𝑐(𝐽 𝑋, 𝑌 ) = 𝑅𝑐(𝐽 2 𝑋, 𝐽 𝑌 ) = −𝑅𝑐(𝑋, 𝐽 𝑌 ) = −𝑅𝑐(𝐽 𝑌 , 𝑋)
= −𝜌(𝑌 , 𝑋).
To prove the other claims, note that Lemma 8.28 implies 𝜌 has the following ex-
pressions in local holomorphic coordinates:
(8.32) 𝜌 = 𝑖𝑅𝑗𝑘 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑘 = −𝑖𝜕𝑗 𝜕𝑘 log(det 𝑔)𝑑𝑧𝑗 ∧ 𝑑𝑧𝑘 = −𝑖𝜕𝜕 log(det 𝑔).
This is a (1, 1)-form, and because 𝜕 ∘ 𝜕 = 𝑑 ∘ 𝜕, this shows that 𝜌 is locally exact and
thus closed. □

The significance of the Ricci form is based on the following theorem.

Theorem 8.31. On a Kähler manifold 𝑀, the Ricci form is equal to 2𝜋 times the
first Chern form of the Chern connection on 𝑇 ′ 𝑀.

Proof. Let ∇′ denote the Chern connection on 𝑇 ′ 𝑀 with respect to the Hermitian
fiber metric ⟨⋅, ⋅⟩ defined by (8.4). Recall from Theorem 8.10(f) that ∇′ is equal to
the restriction of the Levi-Civita connection ∇. To compute its first Chern form, we
work in holomorphic coordinates (𝑧1 , … , 𝑧𝑛 ) and note that the connection forms 𝜃𝑗𝑘

𝜃𝑘𝑙 (𝑋)𝜕𝑙 = ∇′𝑋 𝜕𝑘 = Γ𝑗𝑘 𝑋 𝑗 𝜕𝑙 ,

are determined by
𝑙

𝜃𝑘𝑙 = Γ𝑗𝑘 𝑑𝑧𝑗 .

which implies
𝑙

From formula (7.12), the Chern form is determined in this coordinate domain

𝑖 𝑖
by
𝑑𝜃 𝑙 =
𝑐1 (∇′ ) = 𝑑 Γ𝑙 𝑑𝑧𝑗 ),
2𝜋 𝑙 2𝜋 ( 𝑗𝑙

𝑖 𝑖 1
and (8.30) and (8.32) show that we can rewrite this as

𝑐1 (∇′ ) = 𝑑𝜕 log(det 𝑔) = 𝜕𝜕 log(det 𝑔) = 𝜌.

2𝜋 2𝜋 2𝜋
□

[Cal57] that if (𝑀, 𝑔) is a compact Kähler manifold and 𝜌 ̃ is any closed real (1, 1)-
The Italian-American mathematician Eugenio Calabi conjectured in the 1950s

form representing the cohomology class 2𝜋𝑐1ℝ (𝑇 ′ 𝑀), then there is a Kähler metric
in the same Kähler class whose Ricci form is equal to 𝜌.̃ The conjecture was proved
in 1977 by Shing-Tung Yau [Yau78].
248 8. Hermitian and Kähler Manifolds

Theorem 8.32 (Calabi–Yau). Let (𝑀, 𝑔) be a compact Kähler manifold with Käh-
ler form 𝜔. If 𝜌 ̃ is any closed real (1, 1)-form representing 2𝜋𝑐1ℝ (𝑇 ′ 𝑀), then there
is a unique Kähler metric on 𝑀 whose Kähler form is cohomologous to 𝜔 and
whose Ricci form is equal to 𝜌.̃
The proof of this theorem (which was one of the main accomplishments for
which Yau was awarded the Fields Medal) was based on deep ideas in nonlinear
partial differential equations. We do not have the tools to prove the theorem, but it

A Kähler metric in the same Kähler class as 𝜔 is given by a Kähler form 𝜔 ̃=

is worthwhile outlining how to set it up as a PDE problem.

𝜔 + 𝛾, where 𝛾 is an exact real (1, 1)-form. In fact, the global 𝜕𝜕-lemma (which we
will prove in Chapter 9) shows that every exact real (1, 1)-form can be expressed as
𝛾 = 𝑖𝜕𝜕𝑢 for some smooth real-valued function 𝑢. Thus we seek a function 𝑢 such
that 𝜔
̃ = 𝜔 + 𝑖𝜕𝜕𝑢 is a Kähler form with the prescribed Ricci form.

local holomorphic coordinates. The coefficients of the new metric 𝑔 ̃ are given by
To see how the Ricci form changes under such a change in metric, we work in

𝑔 𝑗𝑘
̃ = 𝑔𝑗𝑘 + 𝜕𝑗 𝜕𝑘 𝑢.
We note also that since the given form 𝜌 ̃ and the Ricci form 𝜌 of 𝑔 both represent 2𝜋
times the first Chern class of 𝑇 ′ 𝑀, they are cohomologous, so the global 𝜕𝜕-lemma
shows that there is smooth real function 𝑣 such that 𝜌 ̃ = 𝜌 + 𝑖𝜕𝜕𝑣.
Using (8.32), we see that in coordinates we have
𝜌 = −𝑖𝜕𝜕 log(det 𝑔𝑗𝑘 ),
𝜌 ̃ = −𝑖𝜕𝜕 log(det(𝑔𝑗𝑘 + 𝜕𝑗 𝜕𝑘 𝑢)),
so we need to solve the following system of PDEs for 𝑢:
(8.33) −𝑖𝜕𝜕 log(det(𝑔𝑗𝑘 + 𝜕𝑗 𝜕𝑘 𝑢)) = −𝑖𝜕𝜕 log(det 𝑔𝑗𝑘 ) + 𝑖𝜕𝜕𝑣.
Calabi realized that this can be hugely simplified by just looking for a function 𝑢

log(det(𝑔𝑗𝑘 + 𝜕𝑗 𝜕𝑘 𝑢)) = log(det 𝑔𝑗𝑘 ) − 𝑣,

that satisfies

which obviously implies (8.33). This last equation is equivalent to

det(𝑔𝑗𝑘 + 𝜕𝑗 𝜕𝑘 𝑢)
= 𝑒−𝑣 ,
det(𝑔𝑗𝑘 )
(8.34)

side is the ratio of the two globally defined volume forms: 𝑑𝑉𝑔 ̃/𝑑𝑉𝑔 . Thus the prob-
and in this form the equation actually makes sense globally, because the left-hand

lem has been reduced to solving a single equation for a single unknown function.
Equation (8.34) is of a type called a complex Monge–Ampère equation: a real

of an unknown function, and the complex version is similar but uses only the (1, 1)-
Monge–Ampère equation is any equation involving the determinant of the Hessian

part of the Hessian (or, as in this case, the Hessian plus some known matrix). It
Ricci and Scalar Curvatures 249

is a fully nonlinear second-order partial differential equation, for which Yau had to
develop a substantial array of new techniques.
There is another natural problem on compact Kähler manifolds that is closely
related to the Calabi–Yau theorem. For Riemann surfaces, there is a powerful re-

says every Riemann surface admits a holomorphic covering by ℂℙ1 , ℂ1 , or the unit
sult called the uniformization theorem (see, for example, [FK92, p. 191]), which

disk. An important geometric consequence of this is that every Riemann surface

possesses a complete Kähler metric with constant Gaussian curvature [FK92, Thm.
IV.8.6]. Much effort has been expended in attempting to generalize this theorem
to higher dimensions. The most obvious generalization—finding a Kähler metric

cause Theorem 8.27 shows that only certain manifolds (those covered by ℂℙ𝑛 , ℂ𝑛 ,
of constant holomorphic sectional curvature—is hopeless in higher dimensions be-

or 𝔹2𝑛 ) admit such metrics.

A plausible candidate for generalizing the uniformization theorem to higher
dimensions would be to search for a Kähler–Einstein metric, which is a Kähler
metric whose Ricci tensor is a constant multiple of the metric, or equivalently whose
Ricci form is a constant multiple of the Kähler form. This condition is satisfied by
every metric of constant holomorphic sectional curvature, but also by many other

Certainly a necessary condition for 𝑀 to possess a Kähler–Einstein metric is

Kähler metrics as well, as we will see below.

has Ricci form 𝜌 = 𝜆𝜔 for some constant 𝜆, and 𝜌 represents a positive multiple of
that it must possess a Kähler metric. Beyond that, since a Kähler–Einstein metric

the first real Chern class of 𝑇 ′ 𝑀, another necessary condition is that the first Chern
class must contain a representative (1, 1)-form that is positive (which would be the
case if 𝜆 > 0), zero (for 𝜆 = 0), or negative (for 𝜆 < 0).
The zero case is a direct consequence of the Calabi–Yau theorem: if 𝑐1ℝ (𝑇 ′ 𝑀)
contains the zero form, then there is a Kähler–Einstein metric whose Ricci form is
zero. Around the same time as Yau proved the Calabi conjecture, Yau [Yau78] and
Thierry Aubin [Aub78] independently solved the negative case, using techniques
very similar to those used in the proof of the Calabi–Yau theorem.

Theorem 8.33 (Existence of Negative Kähler–Einstein Metrics). If (𝑀, 𝑔) is a

compact Kähler manifold and 𝑐1ℝ (𝑇 ′ 𝑀) is represented by a negative (1, 1)-form,
then 𝑀 admits a Kähler–Einstein metric with the same Kähler class as 𝑔.

istence of Kähler–Einstein metrics even when 𝑐1ℝ (𝑇 ′ 𝑀) contains a positive (1, 1)-
The positive case is more complicated, because there are obstructions to the ex-

A connected compact Kähler manifold 𝑀 with holomorphically trivial canon-

form.

tion of trivial canonical bundle implies 𝑐1ℝ (𝑇 ′ 𝑀) = 0, and therefore the Calabi–Yau
ical bundle is called a Calabi–Yau manifold. Problem 8-10 shows that the assump-
250 8. Hermitian and Kähler Manifolds

theorem implies that 𝑀 carries a unique Kähler–Einstein metric with zero Ricci
curvature in each Kähler class (such a metric is said to be Ricci flat).
Here are some examples of Calabi–Yau manifolds.

Example 8.34 (Tori as Calabi–Yau Manifolds). Let Λ ⊆ ℂ𝑛 be a lattice and

𝑀 = ℂ𝑛 /Λ the associated 𝑛-dimensional complex torus. The holomorphic (𝑛, 0)-
form 𝑑𝑧1 ∧ ⋯ ∧ 𝑑𝑧𝑛 ∈ Ω𝑛 (ℂ𝑛 ) is invariant under the action of Λ, and thus descends
to a nonvanishing holomorphic section of the canonical bundle of 𝑀. Therefore,

metrics on 𝑀 are the flat metrics obtained from Hermitian inner products on ℂ𝑛 . //
every complex torus is a Calabi–Yau manifold. The corresponding Ricci-flat Kähler

Example 8.35 (K3 Surfaces). A simply connected compact complex 2-manifold

with trivial canonical bundle is called a K3 surface. (The term was coined by André
Weil [Wei14, p. 546] as a play on the name of the mountain peak K2 in Kashmir, in
honor of Ernst Kummer, Erich Kähler, and Kunihiko Kodaira.) Kodaira [Kod64]

20-parameter family of inequivalent holomorphic structures. It was proved in 1983

proved in 1964 that all K3 surfaces are diffeomorphic to each other, but there is a

by Yum-Tong Siu [Siu83] that every K3 surface has a Kähler metric, and thus is a
Calabi–Yau manifold. //

provide a rich source of examples of Calabi–Yau manifolds. Suppose 𝑆 ⊆ ℂℙ𝑛 is

Example 8.36 (Projective Calabi–Yau Hypersurfaces). Projective hypersurfaces

mial of degree 𝑑. The adjunction formula (Problem 4-10) shows that the canoni-
a nonsingular projective algebraic hypersurface defined by a homogeneous polyno-

cal bundle 𝐾𝑆 is isomorphic to the restriction to 𝑆 of 𝐾ℂℙ𝑛 ⊗ 𝐿𝑆 . On the other

hand, Example 3.40 showed that 𝐿𝑆 ≅ 𝐻 𝑑 , while Proposition 4.17 shows that
𝐾ℂℙ𝑛 ≅ 𝐻 −(𝑛+1) . Thus 𝐾𝑆 is trivial exactly when 𝑑 = 𝑛 + 1, and every non-
singular projective algebraic hypersurface in ℂℙ𝑛 of degree 𝑛 + 1 is a Calabi–Yau

surface of degree 𝑛 + 1 in ℂℙ𝑛 (Example 2.39).

manifold. A simple example in each dimension is provided by the Fermat hyper-
//

Calabi–Yau manifolds play a central role in the approach to elementary parti-

only if spacetime is viewed as the 10-dimensional total space of a fiber bundle

cle theory known as superstring theory, because the theory seems to be consistent

whose base has the familiar 4 dimensions and whose model fiber is a Calabi–Yau
threefold (which is, in particular, a 6-dimensional Ricci-flat Riemannian manifold).
The interaction between physicists and mathematicians has led to striking new in-
sights into the structure of Calabi–Yau manifolds, which are worth learning about
if you want to pursue this field. (See [YN11] for an informal description of this
interaction.)
It should be noted that except for the trivial case of tori, there is no known
Calabi–Yau manifold whose Ricci-flat Kähler metric can be written down explicitly.
Even for Calabi–Yau manifolds that are defined algebraically such as the projective
Problems 251

hypersurfaces of Example 8.36, there is no known algebraic construction of their

associated Ricci-flat Kähler metrics.
The definition of Calabi–Yau manifolds we have given is the simplest and per-
haps the most common one, but you will find a number of closely related but in-
equivalent definitions in the literature. Some definitions require in addition that the

tori). Other definitions relax one or more of the conditions, such as allowing 𝑀
manifold be simply connected or have finite fundamental group (which rules out

to be noncompact or requiring only trivial first real Chern class instead of trivial
canonical bundle. See also Problem 10-11 for another condition that is sometimes
imposed.

Problems
8-1. Suppose (𝑀, 𝑔) is a Kähler manifold of dimension 𝑛 ≥ 2, and 𝑔 ̃ is a metric
conformal to 𝑔 (i.e., 𝑔 ̃ = 𝑓 𝑔 for some smooth positive function 𝑓 ). Prove
that 𝑔 ̃ is Kähler if and only if 𝑓 is constant.
̃ 𝑔 )̃ and (𝑀, 𝑔) are Riemannian manifolds, a smooth submersion
8-2. If (𝑀,
𝜋∶ 𝑀 ̃ → 𝑀 is called a Riemannian submersion if for each 𝑥 ∈ 𝑀, ̃
the total derivative 𝐷𝜋(𝑥) ∶ 𝑇𝑥 𝑀 ̃ → 𝑇𝜋(𝑥) 𝑀 restricts to a linear isom-
etry from 𝐻𝑥 to 𝑇𝜋(𝑥) 𝑀, where 𝐻𝑥 ⊆ 𝑇𝑥 𝑀, ̃ called the horizontal
tangent space, denotes the orthogonal complement of Ker 𝐷𝜋(𝑥). Let
𝜋 ∶ 𝕊2𝑛+1 → ℂℙ𝑛 denote the restriction of the canonical quotient map
ℂ𝑛+1 ∖ {0} → ℂℙ𝑛 . Show that 𝜋 is a Riemannian submersion when 𝕊2𝑛+1
is given the standard round metric and ℂℙ𝑛 is given the Fubini–Study met-
ric. [Hint: Since the metrics on both 𝕊2𝑛+1 and ℂℙ𝑛 are invariant under
U(𝑛 + 1), it suffices to check the condition at one point of 𝕊2𝑛+1 .]

8-3. Let 𝑔 be a Kähler metric on a Riemann surface 𝑀. Show that the holo-
morphic sectional curvature of 𝑔 is equal to its Gaussian curvature, and
in terms of any local holomorphic coordinate 𝑧, both are given by the
formula

1 𝜕2
− log 𝑢,
𝑢 𝜕𝑧𝜕𝑧

where 𝑢 = 𝑔(𝜕/𝜕𝑧, 𝜕/𝜕𝑧).

8-4. Let 𝑄 ⊆ ℂℙ2 be the quadric curve defined by the homogeneous polyno-
mial 𝑤1 𝑤2 − (𝑤0 )2 . Compute the Gaussian curvature of 𝑄 in the metric
obtained by restricting the Fubini-Study metric to 𝑄.
252 8. Hermitian and Kähler Manifolds

8-5. Let 𝑔CH be the complex hyperbolic metric on 𝔹2𝑛 , defined by the Kähler
form 𝜔CH = − 2𝑖 𝜕𝜕 log(1 − |𝑧|2 ) (see Example 8.26(c)). Let U(𝑛, 1) be
the subgroup of GL(𝑛 + 1, ℂ) leaving invariant the following sesquilinear
form:
𝐻(𝑣, 𝑤) = 𝑣1 𝑤1 + ⋯ + 𝑣𝑛 𝑤𝑛 − 𝑣0 𝑤0 .
(a) Considering the unit ball 𝔹2𝑛 as a subset of ℂℙ𝑛 via the embedding
(𝑧1 , … , 𝑧𝑛 ) ↦ [1, 𝑧1 , … , 𝑧𝑛 ], show that U(𝑛, 1) acts transitively on
𝔹2𝑛 by projective transformations.
(b) Show that 𝑔CH is, up to a constant multiple, the unique U(𝑛, 1)-
invariant Riemannian metric on 𝔹2𝑛 .
(c) Show that 𝑔CH is geodesically complete.
(d) Show that 𝑔CH has constant holomorphic sectional curvature equal
to −4.
8-6. Let (𝑀, 𝑔) be one of the following Kähler manifolds with constant holo-
morphic sectional curvature: (ℂℙ𝑛 , 𝑔FS ), (ℂ𝑛 , 𝑔E ), or (𝔹2𝑛 , 𝑔CH ). Prove
that if 𝑈 ⊆ 𝑀 is a connected open set and 𝜑 ∶ 𝑈 → 𝑀 is a holomorphic
local isometry, then 𝜑 is the restriction of a global holomorphic isometry.
[Hint: Choose 𝑝 ∈ 𝑈 and show that there is a global holomorphic isom-
etry 𝜓 ∶ 𝑀 → 𝑀 such that 𝜓 ∘ 𝜑 fixes 𝑝 and an orthonormal basis for
𝑇𝑝′ 𝑀. Then use the exponential map to show that 𝜓 ∘ 𝜑 is the identity on
𝑈 .]
8-7. Let 𝑀 be a complex manifold of dimension 𝑛, and let 𝑔 be a Kähler metric
on 𝑀 with constant holomorphic sectional curvature 𝑐.
(a) Let 𝑣, 𝑤 ∈ 𝑇𝑥 𝑀 be a pair of orthonormal vectors. Show that the (or-
dinary) sectional curvature of 𝑔 in the direction of the plane spanned
by (𝑣, 𝑤) is given by

sec(𝑣, 𝑤) = 14 𝑐 (1 + 3 ⟨𝑣, 𝐽 𝑤⟩2 ) .

(b) Show that if 𝑛 ≥ 2, then at each point of 𝑀, the (ordinary) sectional

curvatures of 𝑔 take on all values between 14 𝑐 and 𝑐, inclusive. Con-
clude that 𝑀 cannot have constant sectional curvature unless it is
flat.
8-8. Calculate the volume of ℂℙ𝑛 with the Fubini–Study metric.
8-9. Let 𝑚 > 𝑛 and let 𝐹 ∶ ℂℙ𝑛 → ℂℙ𝑚 be the holomorphic embedding of
Problem 3-2. Show that 𝐹 pulls back the Fubini–Study metric of ℂℙ𝑚 to
that of ℂℙ𝑛 .
8-10. Suppose (𝑀, 𝑔) is a Kähler manifold and 𝐾 is its canonical bundle. Show
that 𝑐1ℝ (𝑇 ′ 𝑀) = −𝑐1ℝ (𝐾).
Problems 253

8-11. Let (𝑀, 𝑔) be a Kähler manifold, and let (𝑧𝑗 ) be holomorphic coordinates
satisfying (8.8) at a point 𝑎 ∈ 𝑀. Prove that the components of the cur-
vature tensor at 𝑎 are given by

𝑅𝑗𝑘𝑙𝑚 = −𝜕𝑘 𝜕𝑗 𝑔𝑙𝑚 .

8-12. Suppose 𝑀 is a smooth manifold endowed with a Riemannian metric 𝑔

and an almost complex structure 𝐽 satisfying 𝑔(𝐽 𝑋, 𝐽 𝑌 ) = 𝑔(𝑋, 𝑌 ) for
all vector fields 𝑋 and 𝑌 . Suppose in addition that 𝐽 is parallel, that is,
∇𝐽 ≡ 0. Prove that 𝐽 is integrable and 𝑔 is Kähler.
8-13. Let (𝑀, 𝑔) be a Kähler manifold with constant holomorphic sectional cur-
vature 𝑐, and let 𝑁 ⊆ 𝑀 be a complex submanifold endowed with the
induced metric. Show that all holomorphic sectional curvatures of 𝑁 are
less than or equal to 𝑐. [Hint: Use the Gauss equation, [LeeRM, Thm.
8.5].]

folds. Given a Kähler manifold (𝑀, 𝑔), for any two linearly independent
8-14. Here is another curvature quantity that can be defined on Kähler mani-

vectors 𝑍, 𝑊 ∈ 𝑇𝑥′ 𝑀 at the same point 𝑥 ∈ 𝑀, define the holomorphic

bisectional curvature associated with 𝑍, 𝑊 by

𝑅𝑚(𝑍, 𝑍, 𝑊 , 𝑊 )
𝐵(𝑍, 𝑊 ) = .
|𝑍|2 |𝑊 |2

(a) Show that 𝐵(𝑍, 𝑊 ) is equal to the sum of two sectional curvatures:

𝐵(𝑍, 𝑊 ) = sec(Re 𝑍, Im 𝑍) + sec(Re 𝑍, Im 𝑊 ).

(b) Show that if 𝑔 has constant holomorphic sectional curvature 𝑐, then

between 21 𝑐 and 𝑐, inclusive.

its holomorphic bisectional curvatures at each point take on all values

(c) Show that if 𝑔 has positive holomorphic bisectional curvature, then

it has positive Ricci curvature, and the same is true with “positive”
replaced by “nonpositive,” “negative,” or “nonnegative.”
8-15. Let 𝑀 be a compact, connected Riemann surface endowed with a Kähler
metric, and let ∇′ be the Chern connection on 𝑇 ′ 𝑀.
(a) Show that the Chern form 𝑐1 (∇′ ) is equal to 2𝜋
1
𝑘 𝑑𝐴, where 𝑘 is the
Gaussian curvature (see Problem 8-3) and 𝑑𝐴 is the Riemannian area
form.

bundle 𝑇 ′ 𝑀 is 2 − 2𝑔, and the degree of the canonical bundle 𝐾𝑀

(b) Use the Gauss–Bonnet theorem to show that the degree of the line

is 2𝑔 − 2, where 𝑔 is the genus of 𝑀.

254 8. Hermitian and Kähler Manifolds

8-16. Suppose 𝑀 ⊆ ℂℙ2 is a connected nonsingular algebraic curve of genus

𝑔 defined by a single polynomial of degree 𝑑. Prove the genus-degree

(𝑑 − 1)(𝑑 − 2)
formula:
𝑔= .
2
[Hint: Use the results of Propositions 7.30 and 4.17, and Problems 4-10

8-17. Let 𝑀 ⊆ ℂℙ2 be a 1-dimensional complex submanifold defined by a

and 8-15.]

homogeneous polynomial of degree 𝑑, endowed with the restriction of

the Fubini–Study metric. Show that the area of 𝑀 is equal to 𝜋𝑑.
8-18. A Fano manifold is a compact complex manifold whose anticanonical
bundle is positive. Use the Calabi–Yau theorem to prove that a compact
complex manifold is a Fano manifold if and only if it admits a Kähler

8-19. Let (𝑀, 𝑔) be a connected Riemannian 𝑁-manifold and 𝑝 ∈ 𝑀. For any

metric with strictly positive Ricci curvature.

piecewise smooth loop 𝛾 ∶ [𝑎, 𝑏] → 𝑀 based at 𝑝 (meaning 𝛾(𝑎) = 𝛾(𝑏) =

𝑝), let 𝑃 𝛾 ∶ 𝑇𝑝 𝑀 → 𝑇𝑝 𝑀 denote the parallel transport operator along 𝛾
(see [LeeRM, Chap. 4]). The set of all linear maps from 𝑇𝑝 𝑀 to itself

nomy group at 𝒑. Because parallel transport preserves inner products, a

obtained in this way is a group, denoted by Hol(𝑝) and called the holo-

choice of orthonormal basis for 𝑇𝑝 𝑀 yields a representation of Hol(𝑝) as

a subgroup of 𝑂(𝑁).
(a) Show that if 𝑁 = 2𝑛 and 𝑔 is a Kähler metric, then Hol(𝑝) ⊆ U(𝑛)

group of 𝑂(2𝑛) that preserves the inner product and the complex
with respect to an appropriate basis, where we view U(𝑛) as the sub-

structure on ℂ𝑛 , identified with ℝ2𝑛 in the usual way.

(b) Conversely, suppose 𝑔 is Riemannian and Hol(𝑝) ⊆ U(𝑛) for some
𝑝 ∈ 𝑀 and some choice of orthonormal basis for 𝑇𝑝 𝑀. Show that
𝑀 has an integrable almost complex structure with respect to which
𝑔 is Kähler.
 10.1090/gsm/244/09

Chapter 9

Hodge Theory

We have seen in Chapter 6 that sheaf cohomology groups can play the role of ob-
structions to surjectivity of certain maps of global sections of sheaves. And thanks
to the de Rham–Weil theorem, we can often identify sheaf cohomology groups
with the cohomology groups of certain complexes of global sections, such as the
de Rham complex or the Dolbeault complex. But these cohomology groups are
still typically quotients of infinite-dimensional spaces by infinite-dimensional sub-
spaces, so by themselves they are not very practical for computations. Of course,
the de Rham groups are isomorphic to singular cohomology groups, for which there
are a multitude of computational methods coming from algebraic topology. But for
the Dolbeault groups we do not have such topological tools.
The computations would become more tractable if we could single out a partic-
ular representative for each cohomology class, one that has special properties that
might lead to computational simplifications or new insights. A plausible approach
to finding such a representative would be to seek a representative that is “smallest”
or “most efficient” according to some scheme for measuring sizes.
In this chapter, we introduce an inner product and associated norm on the space
of global differential forms on a compact Riemannian manifold, and show that a
closed form minimizes the norm within its cohomology class if and only if it sat-
isfies a certain differential equation; forms satisfying this equation are called har-
monic forms. The main result about harmonic forms is the Hodge theorem, which
says that on a compact Riemannian manifold, every cohomology class has a unique
harmonic representative. The proof of this important theorem is based on a funda-
mental result about elliptic partial differential operators on compact manifolds; the
proof of that result would take us too far afield into the realm of PDE theory, so we
merely state it and give several references where proofs can be found.

255
256 9. Hodge Theory

Then we turn to the case of Hermitian complex manifolds, and prove a simi-
lar Hodge theorem for the Dolbeault complex. The theory takes its most powerful
form on a compact Kähler manifold, where there is a close relationship between
harmonic representatives of de Rham cohomology classes and harmonic represen-
tatives of Dolbeault classes, leading to deep consequences for de Rham cohomology
and sheaf cohomology.
The general strategy of using harmonic forms to deduce topological, geomet-
ric, or complex-analytic properties of manifolds is now known as Hodge theory,
after William V. D. Hodge, who developed the theory in a series of papers in the
1930s, summarized in [Hod41]. It is now one of the most fundamental tools in both
differential and algebraic geometry.

The Hodge Inner Product

To introduce the Hodge theorem in the simplest possible context, we begin by taking
a detour into the realm of Riemannian manifolds. Once the basic results of the the-
ory have been established, we will return below to the case of complex manifolds.
(The work we do in the Riemannian context will not be wasted, though, because the

Suppose (𝑀, 𝑔) is a Riemannian manifold of dimension 𝑁. The Riemannian

Riemannian Hodge theorem also has important applications to complex manifolds.)

metric 𝑔 together with the musical isomorphism ♯ ∶ 𝑇ℂ∗ 𝑀 → 𝑇ℂ 𝑀 yields a Her-

mitian fiber metric ⟨⋅, ⋅⟩ on the space of complex-valued 1-forms defined by

⟨𝛼, 𝛽⟩ = ⟨𝛼 ♯ , 𝛽 ♯ ⟩ = 𝑔 (𝛼 ♯ , 𝛽 ♯ ).

In any smooth coordinates (𝑥1 , … , 𝑥𝑁 ), it is given by

⟨𝑑𝑥𝑗 , 𝑑𝑥𝑘 ⟩ = 𝑔 𝑗𝑘 .

If (𝐸1 , … , 𝐸𝑁 ) is a local orthonormal frame for 𝑇ℂ 𝑀, then its dual coframe

(𝜀1 , … , 𝜀𝑁 ) is a local orthonormal frame for 𝑇ℂ∗ 𝑀.

0 ≤ 𝑞 ≤ 𝑁, we define a Hermitian fiber metric, called the pointwise Hodge in-

We will extend this fiber metric to differential forms of all degrees. For

ner product, on the bundle of complex-valued 𝑞-forms as follows: For 0-forms

(complex-valued functions), we simply set ⟨𝑢, 𝑣⟩ = 𝑢𝑣. For higher-degree forms,
for any local orthonormal coframe (𝜀1 , … , 𝜀𝑁 ) for 𝑇ℂ∗ 𝑀, let ⟨⋅, ⋅⟩ be the pointwise
Hermitian inner product on Λℂ 𝑀 for which the following collection of 𝑞-forms is
𝑞

orthonormal:

{𝜀 1 ∧ ⋯ ∧ 𝜀 𝑞 ∶ 𝑗 1 < ⋯ < 𝑗 𝑞 }.
𝑗 𝑗

Thanks to the following lemma, this yields a well-defined global fiber metric.
The Hodge Inner Product 257

Lemma 9.1. For 𝑞 ≥ 1, the pointwise Hodge inner product ⟨⋅, ⋅⟩ on 𝑞-forms is
uniquely determined by the following condition: for any locally defined 1-forms
𝛼 1 , … , 𝛼 𝑞 , 𝛽 1 , … , 𝛽 𝑞 , we have

(9.1) ⟨𝛼 1 ∧ ⋯ ∧ 𝛼 𝑞 , 𝛽 1 ∧ ⋯ ∧ 𝛽 𝑞 ⟩ = det (⟨𝛼 𝑗 , 𝛽 𝑘 ⟩).

Thus it is well defined, independently of the choice of local coframe.

Proof. Because both sides of (9.1) are linear over 𝐶 ∞ (𝑀; ℂ) in each 𝛼 𝑖 and
conjugate-linear in each 𝛽 𝑖 , it suffices to check the equality for basis covectors
𝛼 𝑖 = 𝜀𝑗𝑖 and 𝛽 𝑖 = 𝜀𝑘𝑖 . Let 𝐽 and 𝐾 denote the multi-indices (𝑗1 , … , 𝑗𝑞 ) and
(𝑘1 , … , 𝑘𝑞 ), respectively, and let 𝑆𝑞 denote the group of permutations of the set
{1, … , 𝑞}.
For the left-hand side of (9.1), since 𝜀𝑗1 ∧ ⋯ ∧ 𝜀𝑗𝑞 and 𝜀𝑘1 ∧ ⋯ ∧ 𝜀𝑘𝑞 match the
basis forms we have declared to be orthonormal up to sign, we have
sgn 𝜎 if 𝐾 = 𝜎(𝐽 ) for some 𝜎 ∈ 𝑆𝑞 ;
⟨𝜀𝑗1 ∧ ⋯ ∧ 𝜀𝑗𝑞 , 𝜀𝑘1 ∧ ⋯ ∧ 𝜀𝑘𝑞 ⟩ =
{0 if 𝐾 is not a permutation of 𝐽 .
(Here sgn 𝜎 denotes the sign of 𝝈, which is equal to 1 if 𝜎 is an even permutation
and −1 if it is odd.)
On the other hand, if 𝐾 = 𝜎(𝐽 ), then the matrix on the right-hand side of (9.1)
is obtained from the identity by applying the permutation 𝜎 to the columns, so its
determinant is equal to the sign of 𝜎. If 𝐾 is not a permutation of 𝐽 , then the matrix
has a column of zeros, so its determinant is zero. □

complex 𝑞-forms, obtained by treating it as a subbundle of a tensor bundle (see

It should be noted that there is another fiber metric on the bundle of real or

[LeeRM, Prop. 2.40]). These inner products are the same for 1-forms, but for 𝑞 >
1, the pointwise Hodge inner product differs from the tensor inner product by a
constant factor depending on 𝑞. (The exact constants depend on which convention
for the wedge product is in use; see [LeeRM, Problem 2-17 on p. 49].) In this book,

When 𝑀 is compact, we can use this pointwise inner product to define a global
we use only the pointwise Hodge inner product for differential forms.

inner product on the space ℰ 𝑞 (𝑀) of smooth complex-valued 𝑞-forms, called the
(global) Hodge inner product, by

(𝛼, 𝛽) = ⟨𝛼, 𝛽⟩𝑑𝑉𝑔 ,

∫𝑀
where we interpret 𝑑𝑉𝑔 to be the Riemannian volume form of 𝑔 if 𝑀 is oriented, and

defined complex number because we are assuming 𝑀 is compact, and it is easy to

otherwise it is the Riemannian density [LeeSM, Prop. 16.45]. This is always a well-

check that it is a Hermitian inner product on the complex vector space ℰ 𝑞 (𝑀). We
can also extend this definition to noncompact 𝑀 if we restrict attention to compactly
258 9. Hodge Theory

supported forms. (This is sometimes called the 𝑳2 inner product, because it is

used to define the Hilbert space 𝐿2 (𝑀; Λℂ 𝑀 ) of square-integrable forms with
𝑞

measurable coefficients. Note, however, that ℰ 𝑞 (𝑀) is not complete under this
inner product, so the tools of Hilbert space theory, such as orthogonal projections

Hodge inner product by (𝛼, 𝛽) with parentheses, and its associated norm by ‖𝛼‖ =
and orthonormal bases, cannot be used here.) We will consistently denote the global

(𝛼, 𝛼)1/2 , reserving the notations ⟨⋅, ⋅⟩ and | ⋅ | for the pointwise inner product and
norm, respectively.
Using this inner product, we can single out a candidate for a “best” representa-
tive of each cohomology class.
Proposition 9.2. On a compact Riemannian manifold 𝑀, a closed 𝑞-form 𝛼 min-
imizes the norm ‖𝛼‖ within its cohomology class if and only if it is orthogonal to
the space ℬ 𝑞 (𝑀) of exact forms. If so, it is the unique minimizer in its cohomology
class.

Proof. Suppose 𝛼 ∈ ℰ 𝑞 (𝑀) is closed and orthogonal to ℬ 𝑞 (𝑀). Any other rep-
resentative of the same cohomology class is of the form 𝛼
̃ = 𝛼 + 𝑑𝛽 for some
𝛽 ∈ ℰ 𝑞−1 (𝑀). The hypothesis implies
‖̃
𝛼 ‖2 = (𝛼 + 𝑑𝛽, 𝛼 + 𝑑𝛽) = ‖𝛼‖2 + 2✘✘✘𝑑𝛽)✘ + ‖𝑑𝛽‖2
✘
Re(𝛼,
≥ ‖𝛼‖2 ,
with equality if and only if 𝑑𝛽 = 0 and thus 𝛼
̃ = 𝛼.
Conversely, suppose 𝛼 minimizes the norm within its cohomology class. For
an arbitrary 𝛽 ∈ ℰ 𝑞−1 (𝑀), the hypothesis implies ‖𝛼 +𝑡𝑑𝛽‖2 ≥ ‖𝛼‖2 for all 𝑡 ∈ ℝ.
Since the smooth real-valued function 𝑡 ↦ ‖𝛼 + 𝑡𝑑𝛽‖2 takes a minimum at 𝑡 = 0,

𝑑
we conclude that

0= ‖𝛼 + 𝑡𝑑𝛽‖2
𝑑𝑡 |𝑡=0
𝑑
= ‖𝛼‖2 + 2𝑡 Re(𝛼, 𝑑𝛽) + 𝑡2 ‖𝛽‖2 )
𝑑𝑡 |𝑡=0 (
= 2 Re(𝛼, 𝑑𝛽).
Applying the same computation with 𝑖𝛽 in place of 𝛽 shows that Im(𝛼, 𝑑𝛽) =
Re(𝛼, 𝑖𝑑𝛽) = 0 as well, so 𝛼 is orthogonal to 𝑑𝛽. □

forms are orthogonal to the image of 𝑑. As a warm-up, let us examine a simpler

To make use of this observation, we need a practical way to identify which

model problem. Suppose 𝑉 and 𝑊 are finite-dimensional complex vector spaces

endowed with Hermitian inner products, and 𝐴 ∶ 𝑉 → 𝑊 is a linear map. The
adjoint of 𝑨 is the unique linear map 𝐴∗ ∶ 𝑊 → 𝑉 that satisfies
⟨𝐴𝑥, 𝑦⟩ = ⟨𝑥, 𝐴∗ 𝑦⟩ for all 𝑥 ∈ 𝑉 and 𝑦 ∈ 𝑊 .
The Hodge Inner Product 259

In terms of orthonormal bases for 𝑉 and 𝑊 , it is the map whose matrix repre-
sentation is the Hermitian adjoint (transposed conjugate) of that of 𝐴, as you can
check.
Lemma 9.3. In the situation described above,
(Im 𝐴)⟂ = Ker(𝐴∗ ).

Proof. Suppose first that 𝑦 ∈ (Im 𝐴)⟂ . Then

|𝐴∗ 𝑦|2 = ⟨𝐴∗ 𝑦, 𝐴∗ 𝑦⟩ = ⟨𝐴𝐴∗ 𝑦, 𝑦⟩ = 0,
which implies 𝐴∗ 𝑦 = 0. Conversely if 𝐴∗ 𝑦 = 0, then for every 𝑥 ∈ 𝑉 we have
⟨𝐴𝑥, 𝑦⟩ = ⟨𝑥, 𝐴∗ 𝑦⟩ = 0,
which implies 𝑦 ⟂ Im 𝐴. □

To apply this to the infinite-dimensional spaces ℰ 𝑞 (𝑀), we first need to show

that operators like 𝑑 have suitable adjoint operators.
Suppose 𝑀 is smooth manifold, and 𝐸 and 𝐹 are smooth vector bundles over
𝑀. (As usual in this book, we assume complex vector bundles, but the theory
works equally well for real ones.) A complex-linear map 𝑃 ∶ Γ(𝐸) → Γ(𝐹 ) is
called a linear differential operator of order 𝒎 if there is a covering of 𝑀 by open
subsets on which there exist smooth local coordinates (𝑥1 , … , 𝑥𝑁 ) for 𝑀 and local
frames (𝑒𝛼 ) for 𝐸 and (𝑓𝛽 ) for 𝐹 , such that for each such subset 𝑈 , the restriction
of 𝑃 𝑢 to 𝑈 has a local expression of the form
𝑚
𝛽𝑗1 …𝑗𝑞
𝑃( 𝑢𝛼 𝑒 𝛼 ) = 𝑝𝛼 𝜕𝑗1 ⋯ 𝜕𝑗𝑞 (𝑢𝛼 )𝑓𝛽
∑ ∑ ∑ ∑
(9.2)
𝛼 𝑞=0 1≤𝑗𝑖 ≤𝑁 𝛼,𝛽

𝛽𝑗1 …𝑗𝑞
for some smooth functions 𝑝𝛼 on 𝑈 , with the 𝑞 = 𝑚 coefficients 𝑝𝛼 1 𝑚 not
𝛽𝑗 …𝑗

all zero (to ensure that it is actually an operator of order 𝑚 and not one of lower
order).

a local operator, meaning that if two sections 𝑢 and 𝑣 agree on an open set 𝑉 ⊆ 𝑀,
It is worth noting that this definition implies that a linear differential operator is

then 𝑃 𝑢|𝑉 = 𝑃 𝑣|𝑉 .

Now let 𝑀 be a Riemannian manifold. For any Hermitian vector bundle 𝐸 →
𝑀, we denote by Γ𝑐 (𝐸) the space of smooth compactly supported sections of 𝐸,
and define a global Hermitian inner product on Γ𝑐 (𝐸) by

(𝑢, 𝑣) = ⟨𝑢, 𝑣⟩𝑑𝑉𝑔 .

∫𝑀
Suppose 𝐸 and 𝐹 are Hermitian vector bundles over 𝑀 and 𝑃 ∶ Γ(𝐸) → Γ(𝐹 ) is a
linear differential operator. Another linear differential operator 𝑃 ∗ ∶ Γ(𝐹 ) → Γ(𝐸)
260 9. Hodge Theory

is called a formal adjoint of 𝑷 if it satisfies the following formula for all 𝑢 ∈ Γ𝑐 (𝐸)
and 𝑣 ∈ Γ𝑐 (𝐹 ):
(9.3) (𝑃 𝑢, 𝑣) = (𝑢, 𝑃 ∗ 𝑣).
(The term “formal” comes from functional analysis, where there is a notion of
Hilbert space adjoint of an operator acting on a Hilbert space of sections of a
bundle, which is required to satisfy (9.3) for a much larger class of sections. By
contrast, the formal adjoint is only required to satisfy it for smooth compactly sup-
ported sections. The formal adjoint is the only type of adjoint we will consider in
this book.)
Proposition 9.4 (Existence and Uniqueness of Formal Adjoints). Every linear
differential operator between smooth vector bundles has a unique formal adjoint,
which is a linear differential operator of the same order.

Proof. First, a comment: The proof looks a little daunting because of the notation,

To prove existence, let {𝑈𝑖 }𝑖∈𝐼 be a cover of 𝑀 by coordinate domains on

but at heart it is really nothing but integration by parts.

which 𝑃 has a coordinate expression like (9.2).

First suppose 𝑣 ∈ Γ𝑐 (𝐹 ) is supported in one of the sets 𝑈𝑖 and 𝑢 ∈ Γ𝑐 (𝐸) is
arbitrary. Using local coordinates to identify 𝑈𝑖 with an open subset of ℝ𝑁 , we
compute
𝛽𝑗1 …𝑗𝑞
(𝑃 𝑢, 𝑣) = 𝑝𝛼 (𝜕𝑗1 ⋯ 𝜕𝑗𝑞 𝑢 )𝑣𝛾 ℎ𝛽𝛾 √det 𝑔 𝑑𝑥 ⋯ 𝑑𝑥 ,
𝛼 1 𝑁
∑∑∑∫𝑈
𝑞 𝑗𝑖 𝛼𝛽𝛾 𝑖

where ℎ𝛽𝛾 = ⟨𝑓𝛽 , 𝑓𝛾 ⟩. Since the integrand is smooth and compactly supported in
𝑈𝑖 , we can consider this as an integral over a large cube in ℝ𝑁 and integrate by parts
with respect to each variable in turn. The boundary terms vanish, and we obtain
𝛽𝑗1 …𝑗𝑞
(𝑃 𝑢, 𝑣) = (−1)𝑞 𝑢𝛼 𝜕𝑗𝑞 ⋯ 𝜕𝑗1 (𝑝𝛼 𝑣𝛾 ℎ𝛽𝛾 √det 𝑔 )𝑑𝑥1 ⋯ 𝑑𝑥𝑁 .
∑ ∑∑∫𝑈
𝑞 𝑗𝑖 𝛼𝛽𝛾 𝑖

This is equal to the inner product (𝑢, 𝑄𝑖 𝑣), where

1 𝛽𝑗1 …𝑗𝑞 𝛾
(9.4) 𝑄𝑖 𝑣 = (−1)𝑞 𝜕 ⋯ 𝜕𝑗1 (𝑝𝛼 𝑣 ℎ𝛽𝛾 √det 𝑔 )𝐻 𝜎𝛼 𝑒𝜎 ,
∑
√det 𝑔 𝑞 ∑ ∑ 𝑗𝑞
𝑗𝑖 𝛼𝛽𝛾𝜎

as you can check. (Here 𝐻 𝜎𝛼 denotes the inverse of the matrix 𝐻𝛼𝜎 = ⟨𝑒𝛼 , 𝑒𝜎 ⟩.)
For the general case, let {𝜑𝑖 }𝑖∈𝐼 be a partition of unity subordinate to the 𝑈𝑖 ’s,
and define 𝑄 ∶ Γ(𝐹 ) → Γ(𝐸) by
𝑄𝑣 = 𝑄𝑖 (𝜑𝑖 𝑣).
∑
(9.5)
𝑖∈𝐼
Because the supports of the 𝜑𝑖 ’s are locally finite, each point of 𝑀 has a neigh-
borhood on which only finitely many terms in the sum on the right-hand side are
The Hodge Inner Product 261

nonzero, so this is well defined. After we expand out the derivatives using the

𝛽𝑗1 …𝑗𝑞
ing various derivatives and conjugates of 𝑝𝛼 , ℎ𝛽𝛾 , 𝐻 𝜎𝛼 , det 𝑔, and 𝜑𝑖 ; thus it
product rule, we see that this has the same form as (9.2), with coefficients involv-

is a differential operator of the same order as 𝑃 . Based on the argument in the first
part of the proof, for 𝑢 and 𝑣 compactly supported we compute
(𝑃 𝑢, 𝑣) = (𝑃 𝑢, 𝜑𝑖 𝑣) = (𝑢, 𝑄𝑖 (𝜑𝑖 𝑣)) = (𝑢, 𝑄𝑣)
∑ ∑
𝑖∈𝐼 𝑖∈𝐼
(using the fact that there are now only finitely many values of 𝑖 for which 𝜑𝑖 𝑣 is not
identically zero), so 𝑄 is a formal adjoint of 𝑃 .
To prove uniqueness, suppose 𝑄 and 𝑄 ̃ are both formal adjoints of 𝑃 . Let
̃
𝛿𝑣 = 𝑄𝑣 − 𝑄𝑣 for 𝑣 ∈ Γ(𝐹 ); then by the definition of formal adjoint, for all
𝑢 ∈ Γ𝑐 (𝐸) and 𝑣 ∈ Γ𝑐 (𝐹 ), we have
̃ = (𝑃 𝑢, 𝑣) − (𝑃 𝑢, 𝑣) = 0.
(𝑢, 𝛿𝑣) = (𝑢, 𝑄𝑣) − (𝑢, 𝑄𝑣)
Applying this with 𝑢 = 𝛿𝑣, we see that 𝛿𝑣 = 0 for all compactly supported 𝑣. Since

̃
of each point and differential operators act locally, this implies 𝑄 = 𝑄.
every smooth section agrees with a compactly supported section in a neighborhood

Proposition 9.5 (Properties of Formal Adjoints). Suppose 𝐸, 𝐹 , and 𝐺 are

□

smooth Hermitian vector bundles over a Riemannian manifold 𝑀, and 𝑃 ∶ Γ(𝐸) →

Γ(𝐹 ) and 𝑄 ∶ Γ(𝐹 ) → Γ(𝐺) are linear differential operators with formal adjoints
𝑃 ∗ and 𝑄∗ , respectively.
(a) (𝑃 ∗ )∗ = 𝑃 .
(b) (𝑄 ∘ 𝑃 )∗ = 𝑃 ∗ ∘ 𝑄∗ .

Proof. Part (a) follows from the fact that (𝑃 ∗ 𝑢, 𝑣) = (𝑣, 𝑃 ∗ 𝑢) = (𝑃 𝑣, 𝑢) = (𝑢, 𝑃 𝑣)
for all 𝑢 ∈ Γ𝑐 (𝐹 ) and 𝑣 ∈ Γ𝑐 (𝐸), so 𝑃 satisfies the criterion to be the adjoint of 𝑃 ∗ .
To prove (b), for all 𝑢 ∈ Γ𝑐 (𝐸) and 𝑣 ∈ Γ𝑐 (𝐺), we have
(𝑄 ∘ 𝑃 𝑢, 𝑣) = (𝑃 𝑢, 𝑄∗ 𝑣) = (𝑢, 𝑃 ∗ ∘ 𝑄∗ 𝑣),
which shows that 𝑃 ∗ ∘ 𝑄∗ is the formal adjoint of 𝑄 ∘ 𝑃 . □

Using the concept of formal adjoints, we have the following analogue of Lemma
9.3.
Proposition 9.6. Suppose 𝐸 and 𝐹 are smooth Hermitian vector bundles over a
compact Riemannian manifold 𝑀, and 𝑃 ∶ Γ(𝐸) → Γ(𝐹 ) is a linear differential
operator. Then (Im 𝑃 )⟂ = Ker(𝑃 ∗ ).

Proof. Because all sections of 𝐸 and 𝐹 are compactly supported in this case, the
relation (𝑃 𝑢, 𝑣) = (𝑢, 𝑃 ∗ 𝑣) holds for all 𝑢 ∈ Γ(𝐸) and 𝑣 ∈ Γ(𝐹 ). Based on this
formula, the proof of the proposition is exactly the same as the proof of Lemma
9.3. □
262 9. Hodge Theory

As a consequence of Proposition 9.4, for any linear differential operator 𝑃 ,

we can freely refer to its unique formal adjoint 𝑃 ∗ . In particular, 𝑑 ∗ denotes the
formal adjoint of the exterior derivative operator. The following corollary follows
immediately from Propositions 9.2 and 9.6.
Corollary 9.7. On a compact Riemannian manifold 𝑀, a closed differential form
𝛼 is the unique representative of its de Rham cohomology class that minimizes the
Hodge norm if and only if it satisfies 𝑑 ∗ 𝛼 = 0. □
Although the conclusion of Proposition 9.6 is exactly the same as that of Lemma

In the case of an operator 𝐴 ∶ 𝑉 → 𝑊 between finite-dimensional inner product

9.3, it should be noted that there is one crucial difference between the two situations.

spaces, the fact that (Im 𝐴)⟂ = Ker(𝐴∗ ) implies also that Im 𝐴 = (Ker 𝐴∗ )⟂ , just
by taking orthogonal complements of both sides. In other words, given 𝑦 ∈ 𝑊 ,
the equation 𝐴𝑥 = 𝑦 has a solution if and only if 𝑦 is orthogonal to the kernel
of 𝐴∗ . But in the case of a differential operator 𝑃 ∶ Γ(𝐸) → Γ(𝐹 ), the infinite-
dimensional inner product space Γ(𝐹 ) is not complete, so it need not be the case
that ((Im 𝑃 )⟂ )⟂ = Im 𝑃 . Thus we need to do more work to determine when an
equation like 𝑃 𝑢 = 𝑓 has a solution.

The Hodge Star Operator

To make use of Corollary 9.7, we need to compute the formal adjoint of 𝑑
explicitly. For our purposes, the computations will be easier if we assume 𝑀 is
oriented, which is certainly the case for every complex manifold. The main tool
will be a linear operator defined on every oriented Riemannian manifold, called the
Hodge star operator, that takes differential forms to forms of the complementary

Let 𝑀 be an oriented Riemannian 𝑁-manifold. The Hodge star operator will be

degree.

a smooth bundle homomorphism taking 𝑞-forms to (𝑁 −𝑞)-forms. We define it first

in terms of local coframes. Let (𝜀1 , … , 𝜀𝑁 ) be a local real oriented orthonormal
coframe for 𝑇 ∗ 𝑀; thus the collection of forms 𝜀𝑗1 ∧ ⋯ ∧ 𝜀𝑗𝑞 with 𝑗1 < ⋯ < 𝑗𝑞
forms an orthonormal frame for Λℂ 𝑀. To simplify the notation, given any ordered
𝑞

𝑞-tuple of positive integers 𝐽 = (𝑗1 , … , 𝑗𝑞 ), we will use the abbreviation 𝜀𝐽 for

𝜀𝑗1 ∧ ⋯ ∧ 𝜀𝑗𝑞 .
Given any increasing multi-index 𝐽 , the idea is to define
∗ 𝜀𝐽 = sgn 𝜎(𝐽 , 𝐽 ′ )𝜀𝐽
′
(9.6)
and extend to all complex-valued 𝑞-forms by complex linearity, where 𝐽 ′ is the
unique increasing multi-index of length 𝑁 − 𝑞 consisting of the indices comple-
mentary to {𝑗1 , … , 𝑗𝑞 }, and 𝜎(𝐽 , 𝐽 ′ ) is the permutation that takes (1, … , 𝑁) to
(𝑗1 , … , 𝑗𝑞 , 𝑗1′ , … , 𝑗𝑁−𝑞
′
). The next proposition shows that this is well defined, in-
dependently of the choice of coframe.
The Hodge Inner Product 263

Proposition 9.8. On an oriented Riemannian 𝑁-manifold (𝑀, 𝑔), there is a unique

smooth complex-linear bundle isomorphism ∗ ∶ Λℂ 𝑀 → Λℂ 𝑀, called the
𝑞 𝑁−𝑞

Hodge star operator, that satisfies the following formula for all 𝛼, 𝛽 ∈ Λℂ 𝑀:
𝑞

(9.7) 𝛼 ∧ ∗ 𝛽 = ⟨𝛼, 𝛽⟩𝑑𝑉𝑔 ,

where 𝑑𝑉𝑔 is the Riemannian volume form. In terms of any local real oriented
orthonormal coframe, ∗ is given by (9.6). In addition,

(9.8) ∗ 𝛼 = ∗ 𝛼;
∗ ∗ 𝛼 = (−1)𝑞(𝑁−𝑞) 𝛼;
⟨∗ 𝛼, ∗ 𝛽⟩ = ⟨𝛼, 𝛽⟩.
(9.9)
(9.10)

Proof. We begin by showing that the operator ∗ defined locally by (9.6) in terms of

complex-linear in 𝛼 and conjugate-linear in 𝛽, it suffices to prove it for 𝛼 and 𝛽 equal

an oriented real orthonormal coframe satisfies (9.7). Because both sides of (9.7) are

to basis forms. Thus let 𝛼 = 𝜀𝐽 and 𝛽 = 𝜀𝐾 , where 𝐽 and 𝐾 are arbitrary increasing
multi-indices of length 𝑞. Because we assumed the basis forms are real, we have
𝛽 = 𝛽. Then ∗ 𝛽 = sgn 𝜎(𝐾, 𝐾 ′ )𝜀𝐾 , which implies that 𝛽 ∧ ∗ 𝛽 = 𝜀1 ∧ ⋯ ∧ 𝜀𝑁 =
′

𝑑𝑉𝑔 .
First assume 𝐽 = 𝐾. Then 𝛼 = 𝛽, ⟨𝛼, 𝛽⟩ = 1, and

𝛼 ∧ ∗ 𝛽 = 𝛽 ∧ ∗ 𝛽 = 𝑑𝑉𝑔 = ⟨𝛼, 𝛽⟩𝑑𝑉𝑔 ,

which proves (9.7) in this case. On the other hand, if 𝐽 ≠ 𝐾, then ⟨𝛼, 𝛽⟩ = 0, and
𝛼 ∧ ∗ 𝛽 = ±𝜀𝐽 ∧ 𝜀𝐾 = 0 because there is some index common to 𝐽 and 𝐾 ′ .
′

To show that ∗ is the unique operator satisfying (9.7), suppose ∗̃ also satisfies
the same property, and let 𝛿 = ∗ −∗.̃ Then 𝛼 ∧ 𝛿𝛽 = 0 for all 𝛼 and 𝛽, and applying
this with 𝛽 in place of 𝛽 shows that 𝛼 ∧ 𝛿𝛽 = 0 as well. Thus

|𝛿𝛽|2 𝑑𝑉𝑔 = 𝛿𝛽 ∧ ∗ 𝛿𝛽 = ± ∗ 𝛿𝛽 ∧ 𝛿𝛽 = 0,

which shows that 𝛿 is identically zero. Therefore, the definitions given by different
local coframes agree, so ∗ is globally well defined, and it is smooth because it takes

Now (9.8) follows from the fact that ∗ takes real basis forms to real forms and
a smooth local frame to a smooth local frame.

To prove (9.9), it suffices to consider the case 𝛼 = 𝜀𝐽 for some increasing multi-
is extended by complex linearity.

index 𝐽 . On the one hand, ∗ 𝛼 = ±𝜀𝐽 and thus ∗ ∗ 𝛼 = ±𝜀𝐽 = ±𝛼, with the sign
′

determined uniquely by the requirement that ∗ 𝛼 ∧ ∗ ∗ 𝛼 = 𝑑𝑉𝑔 . On the other hand,

since 𝛼 is a 𝑞-form and ∗ 𝛼 is an (𝑁 − 𝑞)-form,

𝑑𝑉𝑔 = 𝛼 ∧ ∗ 𝛼 = (−1)𝑞(𝑁−𝑞) ∗ 𝛼 ∧ 𝛼,
264 9. Hodge Theory

so (−1)𝑞(𝑁−𝑞) 𝛼 satisfies the sign requirement and thus is equal to ∗ ∗ 𝛼. It also

follows from this formula that ∗ is a bundle isomorphism, because ± ∗ is an inverse

Finally, (9.6) shows that ∗ takes an orthonormal frame for Λℂ 𝑀 to an orthonor-

for it.
𝑞

mal frame for Λℂ 𝑀, so (9.10) follows.

𝑁−𝑞
□

For forms of degrees 0 and 1, there are simple formulas for the Hodge star
operator.
Proposition 9.9. On an oriented Riemannian manifold (𝑀, 𝑔), the following for-
mulas hold for a 0-form (scalar function) 𝑢 and a 1-form 𝛽:
(9.11) ∗ 𝑢 = 𝑢 𝑑𝑉𝑔 ,
(9.12) ∗ 𝛽 = 𝛽 ♯ ⅃ 𝑑𝑉𝑔 .

suitable forms 𝛼 and 𝛽. For scalar functions 𝑢 and 𝑣,

Proof. We just need to verify that the given formulas satisfy condition (9.7) for all

𝑣 ∧ (𝑢 𝑑𝑉𝑔 ) = (𝑣𝑢) 𝑑𝑉𝑔 = ⟨𝑣, 𝑢⟩𝑑𝑉𝑔 ,

which proves (9.11). For 1-forms 𝛼 and 𝛽, we use the fact that interior multiplication
by a vector is an antiderivation. Because 𝛼 ∧ 𝑑𝑉𝑔 is an (𝑁 + 1)-form on an 𝑁-
manifold, it is zero, so

0 = 𝛽 ♯ ⅃ (𝛼 ∧ 𝑑𝑉𝑔 ) = (𝛽 ♯ ⅃ 𝛼 )𝑑𝑉𝑔 − 𝛼 ∧ (𝛽 ♯ ⅃ 𝑑𝑉𝑔 )

= ⟨𝛼, 𝛽⟩𝑑𝑉𝑔 − 𝛼 ∧ (𝛽 ♯ ⅃ 𝑑𝑉𝑔 ),

which proves (9.12). □

Using (9.7), we can write the global Hodge inner product as

(𝛼, 𝛽) = 𝛼 ∧ ∗ 𝛽.
∫
(9.13)
𝑀
This is the formula we will use to determine the formal adjoint of 𝑑.
Proposition 9.10. For an oriented Riemannian 𝑁-manifold 𝑀, the formal adjoint
of the exterior derivative operator 𝑑 ∶ ℰ 𝑞−1 (𝑀) → ℰ 𝑞 (𝑀) is the differential op-
erator 𝑑 ∗ ∶ ℰ 𝑞 (𝑀) → ℰ 𝑞−1 (𝑀) given by
(9.14) 𝑑 ∗ 𝛽 = (−1)𝑁𝑞+𝑁+1 ∗ 𝑑 ∗ 𝛽.

Proof. Let 𝛿𝛽 denote the right-hand side of (9.14). The proposition will be proved
if we can show that the following holds for all compactly supported 𝛼 ∈ ℰ 𝑞−1 (𝑀)
and 𝛽 ∈ ℰ 𝑞 (𝑀):
(9.15) (𝛼, 𝛿𝛽) = (𝑑𝛼, 𝛽).
Elliptic Differential Operators 265

Using (9.13), we have

(𝛼, 𝛿𝛽) = 𝛼 ∧ ∗ 𝛿𝛽
∫𝑀

= (−1)𝑁𝑞+𝑁+1 𝛼 ∧ ∗ ( ∗ 𝑑 ∗ 𝛽)
∫𝑀

= (−1)𝑁𝑞+𝑁+1 (−1)(𝑁−𝑞+1)(𝑞−1) 𝛼 ∧ 𝑑 ∗𝛽
∫𝑀

= (−1)𝑞 𝛼 ∧ 𝑑 ∗ 𝛽,
∫𝑀

where in the third line we have used (9.9) together with the fact that 𝑑 ∗ 𝛽 is an
(𝑁 − 𝑞 + 1)-form. Stokes’s theorem gives

0= 𝑑 (𝛼 ∧ ∗ 𝛽 )
∫𝑀

= 𝑑𝛼 ∧ ∗ 𝛽 + (−1)𝑞−1 𝛼 ∧ 𝑑 ∗𝛽
∫𝑀 ∫𝑀
= (𝑑𝛼, 𝛽) − (𝛼, 𝛿𝛽). □

Elliptic Differential Operators

We are going to prove that every de Rham cohomology class on a compact oriented
Riemannian manifold has a unique representative that minimizes the Hodge norm
in its cohomology class. To get started, we show how to convert this to a problem

Starting with a closed 𝑞-form 𝛼 ∈ ℰ 𝑞 (𝑀), we seek a (𝑞 − 1)-form 𝜂 such that

in partial differential equations.

𝛼 − 𝑑𝜂 satisfies 𝑑 ∗ (𝛼 − 𝑑𝜂) = 0. Thus we need to solve the partial differential

equation 𝑑 ∗ 𝑑𝜂 = 𝑑 ∗ 𝛼 for the unknown form 𝜂.
Let us see how this works in the special case 𝑞 = 1. In this case 𝑑 ∗ 𝛼 is a smooth
scalar function, and we are looking for a function 𝑢 such that 𝑑 ∗ 𝑑𝑢 = 𝑑 ∗ 𝛼. This
operator 𝑑 ∗ 𝑑 on functions has another guise. Recall that the gradient of a function
𝑢 on a Riemannian manifold is the vector field grad 𝑢 defined by

grad 𝑢 = (𝑑𝑢)♯ ,
and the divergence of a vector field 𝑋 is the unique scalar function div 𝑋 that sat-
isfies
𝑑(𝑋 ⅃ 𝑑𝑉𝑔 ) = (div 𝑋)𝑑𝑉𝑔 .
The Laplace–Beltrami operator is the partial differential operator Δ ∶ ℰ (𝑀) →
ℰ (𝑀) defined by
Δ𝑢 = div grad 𝑢.
266 9. Hodge Theory

(Some authors define the Laplace–Beltrami operator to be the negative of this, so

as to eliminate the negative sign in the following lemma. But the definition above
is most common among Riemannian geometers.)
Lemma 9.11. Let (𝑀, 𝑔) be a Riemannian manifold. The operator 𝑑 ∗ 𝑑 acting on
scalar functions is equal to the negative of the Laplace–Beltrami operator: 𝑑 ∗ 𝑑𝑢 =
−Δ𝑢.

Proof. By Proposition 9.10, 𝑑 ∗ 𝑑𝑢 = − ∗ 𝑑 ∗ 𝑑𝑢. Using Proposition 9.9, we com-

pute
𝑑 ∗ 𝑑𝑢 = − ∗ 𝑑 ((𝑑𝑢)♯ ⅃ 𝑑𝑉𝑔 ) = − ∗ ((div grad 𝑢) 𝑑𝑉𝑔 ) = − div grad 𝑢. □

Thus to find a norm-minimizing representative of the cohomology class of 𝛼,

we need to solve the equation Δ𝑢 = −𝑑 ∗ 𝛼 for 𝑢. There is a necessary condition
for the equation Δ𝑢 = 𝑓 to have a solution on a compact manifold 𝑀: if 𝑀0 is
any connected component of 𝑀, then by the divergence theorem, ∫𝑀 (Δ𝑢)𝑑𝑉𝑔 =
∫𝑀 div(grad 𝑢)𝑑𝑉𝑔 = 0 since 𝜕𝑀0 is empty. But this is the only obstruction. The
0

0
essential fact about the Laplace–Beltrami operator is the following theorem.
Theorem 9.12. On a compact oriented Riemannian manifold 𝑀, the equation
Δ𝑢 = 𝑓 has a solution 𝑢 if and only if the integral of 𝑓 over each connected com-
ponent of 𝑀 is zero.

about differential operators. Accepting this for now, let 𝑀0 be any connected com-
We will prove this below, as a consequence of a much more general theorem

ponent of 𝑀 and let 𝜒0 ∈ 𝐶 ∞ (𝑀) be the function that is equal to 1 on 𝑀0 and 0

on all the other components of 𝑀. Then since 𝜒0 is locally constant,

(𝑑 ∗ 𝛼)𝑑𝑉𝑔 = (𝑑 ∗ 𝛼)𝜒 0 𝑑𝑉𝑔 = (𝑑 ∗ 𝛼, 𝜒0 ) = (𝛼, 𝑑𝜒0 ) = 0,

∫𝑀0 ∫𝑀
so 𝑑 ∗ 𝛼 satisfies the necessary and sufficient condition for the existence of a function
𝑢 satisfying Δ𝑢 = −𝑑 ∗ 𝛼 and thus 𝑑 ∗ (𝛼 − 𝑑𝑢) = 0. The 1-form 𝛼 − 𝑑𝑢 is then the
(necessarily unique) norm-minimizing representative of the cohomology class [𝛼].
The feature of the Laplace–Beltrami operator that makes a characterization like
that of Theorem 9.12 possible is that it is elliptic, a condition we will define shortly.

bol of a differential operator. To begin, for a linear differential operator 𝑃 of

To define ellipticity, we need to introduce the notion of the principal sym-

order 𝑚, expressed in coordinates as in (9.2), we define the total symbol of 𝑷

in these coordinates to be the matrix-valued function 𝑝(𝑥, 𝜉) of 2𝑁 variables
(𝑥1 , … , 𝑥𝑁 , 𝜉1 , … , 𝜉𝑁 ) obtained by replacing every occurrence of 𝜕𝑗 in the formula
for 𝑃 with a new independent variable 𝜉𝑗 :
𝑚
𝛽𝑗1 …𝑗𝑞
𝑝𝛼 (𝑥, 𝜉) = 𝑝𝛼 (𝑥)𝜉𝑗1 ⋯ 𝜉𝑗𝑞 .
𝛽
∑ ∑
𝑞=0 1≤𝑗𝑖 ≤𝑁
Elliptic Differential Operators 267

The principal symbol is the matrix-valued function 𝜎𝑃 (𝑥, 𝜉) obtained by keeping

only the terms of highest order 𝑞 = 𝑚 in 𝜉:

(𝜎𝑃 )𝛼 (𝑥, 𝜉) = 𝑝𝛼 (𝑥)𝜉𝑗1 ⋯ 𝜉𝑗𝑚 .

𝛽 𝛽𝑗1 …𝑗𝑚
∑
(9.16)
1≤𝑗𝑖 ≤𝑁

Each matrix entry is a homogeneous polynomial of degree 𝑚 in (𝜉1 , … , 𝜉𝑁 ), with

coefficients that are smooth functions of 𝑥.
When we change to different coordinates or different local frames, the total
symbol transforms by a complicated formula involving derivatives of the coeffi-
cients and of the coordinate transition functions, so it is of limited use on mani-

Recall that if (𝑥1 , … , 𝑥𝑁 ) are smooth local coordinates for 𝑀, the natural coordi-
folds. However, the principal symbol has a coordinate-independent interpretation.

nates for 𝑻 ∗ 𝑴 are the coordinates (𝑥, 𝜉) = (𝑥1 , … , 𝑥𝑁 , 𝜉1 , … , 𝜉𝑁 ) defined by the

correspondence

𝜉𝑗 𝑑𝑥𝑗 |𝑝 ↔ (𝑥1 (𝑝), … , 𝑥𝑁 (𝑝), 𝜉1 , … , 𝜉𝑁 ).

(See [LeeSM, p. 277].)

Proposition 9.13. Suppose 𝐸 and 𝐹 are smooth complex vector bundles over a
smooth manifold 𝑀, and 𝑃 ∶ Γ(𝐸) → Γ(𝐹 ) is a linear differential operator of
order 𝑚. There is a globally defined smooth map 𝜎𝑃 ∶ 𝑇 ∗ 𝑀 → Hom(𝐸, 𝐹 ) called
the principal symbol of 𝑷 , such that for each (𝑥, 𝜉) ∈ 𝑇 ∗ 𝑀, 𝜎𝑃 (𝑥, 𝜉) is the linear
map from 𝐸𝑥 to 𝐹𝑥 whose matrix representation with respect to local frames for 𝐸
and 𝐹 and natural coordinates for 𝑇 ∗ 𝑀 is given by (9.16). For any point 𝑥0 ∈ 𝑀
and covector 𝜉0 ∈ 𝑇𝑥∗0 𝑀, the linear map 𝜎𝑃 (𝑥0 , 𝜉0 ) ∶ 𝐸𝑥0 → 𝐹𝑥0 satisfies

1
𝜎𝑃 (𝑥0 , 𝜉0 )𝑢0 = 𝑃 (𝑣𝑚 𝑢)(𝑥0 ),
𝑚!
(9.17)

where 𝑢 is any smooth local section of 𝐸 that satisfies 𝑢(𝑥0 ) = 𝑢0 and 𝑣 is any
smooth real-valued function on a neighborhood of 𝑥0 that satisfies 𝑣(𝑥0 ) = 0 and
𝑑𝑣𝑥0 = 𝜉0 .

Proof. It suffices to show that the map whose local matrix representation is defined
by (9.16) also satisfies (9.17), because then (9.17) shows it is independent of choices

Thus given (𝑥0 , 𝜉0 ) ∈ 𝑇 ∗ 𝑀, let (𝑥𝑗 ) be local coordinates for 𝑀 and (𝑒𝛼 ) and
of coordinates and frames, and (9.16) shows that it is smooth.

(𝑓𝛽 ) be local frames, and choose 𝑢 and 𝑣 as in the statement of the proposition.
We can write 𝑢0 = 𝑢(𝑥0 ) = ∑𝛼 𝑢𝛼 (𝑥0 )𝑒𝛼 (𝑥0 ) and 𝜉0 = ∑𝑗 𝜉𝑗 𝑑𝑥𝑗 |𝑥0 where 𝜉𝑗 =
𝜕𝑗 𝑣(𝑥0 ). When we apply 𝑃 to 𝑣𝑚 𝑢 and expand out the derivatives using the product
rule, every term in which fewer than 𝑚 derivatives fall on the 𝑣𝑚 factor will vanish
268 9. Hodge Theory

when we set 𝑥 = 𝑥0 because 𝑣(𝑥0 ) = 0, so the only terms that remain are

𝑃 (𝑣𝑚 𝑢)(𝑥0 ) = 𝑝𝛼 (𝑥0 )(𝜕𝑗1 ⋯ 𝜕𝑗𝑚 (𝑣𝑚 ))|𝑥=𝑥 𝑢𝛼 (𝑥0 )𝑓𝛽 (𝑥0 )
𝛽𝑗1 …𝑗𝑚
∑ ∑ 0
1≤𝑗𝑖 ≤𝑁 𝛼,𝛽

= 𝑚! 𝑝𝛼 (𝑥0 )𝜉𝑗1 ⋯ 𝜉𝑗𝑚 𝑢𝛼 (𝑥0 )𝑓𝛽 (𝑥0 )

𝛽𝑗1 …𝑗𝑚
∑ ∑
1≤𝑗𝑖 ≤𝑁 𝛼,𝛽

= 𝑚! 𝜎𝑃 (𝑥0 , 𝜉0 )𝑢0 . □

homomorphism 𝑃 ∶ Γ(𝐸) → Γ(𝐹 ), in which case the principal symbol of 𝑃 is 𝑃

We note that a linear differential operator of order zero is just a smooth bundle

itself, and does not depend on 𝜉.

A linear differential operator 𝑃 ∶ Γ(𝐸) → Γ(𝐹 ) is said to be elliptic if for every
𝑥 ∈ 𝑀 and every nonzero 𝜉 ∈ 𝑇𝑥∗ 𝑀, the homomorphism 𝜎𝑃 (𝑥, 𝜉) ∶ 𝐸𝑥 → 𝐹𝑥
is invertible. Note that this is possible only if the bundles 𝐸 and 𝐹 have the same
rank. In particular, if 𝑃 ∶ ℰ (𝑀) → ℰ (𝑀) is an operator on scalar functions, we
can think of it as acting on sections of the trivial complex line bundle 𝑀 × ℂ → 𝑀,
so its principal symbol at each point (𝑥, 𝜉) ∈ 𝑇 ∗ 𝑀 is just a linear map from ℂ to
itself, which is multiplication by a complex number. In that case, we can treat 𝜎𝑃
as a complex-valued function on 𝑇 ∗ 𝑀, and 𝑃 is elliptic if and only if this function
is never zero for 𝜉 ≠ 0.
The term “elliptic” may seem mysterious in this context, so it is worth pointing
out where it comes from. The first linear partial differential operators to be studied

scalar functions on ℝ2 . Such an operator has the form

systematically were the constant-coefficient second-order operators acting on real

𝜕2𝑢 𝜕2𝑢 𝜕2𝑢 𝜕𝑢 𝜕𝑢

𝑃𝑢 = 𝐴 + 𝐵 + 𝐶 +𝐷 +𝐸 + 𝐹 𝑢,
𝜕𝑥 2 𝜕𝑥𝜕𝑦 𝜕𝑦 2 𝜕𝑥 𝜕𝑦
for some real constants 𝐴, 𝐵, 𝐶, 𝐷, 𝐸, 𝐹 with 𝐴, 𝐵, 𝐶 not all zero. Because 𝑃 is a
scalar operator, its principal symbol is the scalar-valued function given by

𝜎𝑃 ((𝑥, 𝑦), (𝜉, 𝜂)) = 𝐴𝜉 2 + 𝐵𝜉𝜂 + 𝐶𝜉 2 .

This is a homogeneous quadratic polynomial in (𝜉, 𝜂). It is invertible (as a linear

map on the 1-dimensional fibers) for a particular value of ((𝑥, 𝑦), (𝜉, 𝜂)) if and only
𝜎𝑃 ((𝑥, 𝑦), (𝜉, 𝜂)) ≠ 0. The only homogeneous quadratic polynomials in (𝜉, 𝜂) that
are never zero for (𝜉, 𝜂) ≠ 0 are the ones that are positive-definite or negative-
definite, which can be detected by the discriminant condition 𝐵 2 − 4𝐴𝐶 < 0; thus
𝑃 is elliptic if and only if its principal symbol is positive- or negative-definite. These
are exactly the homogeneous quadratic polynomials whose level curves are ellipses,

linear change of coordinates to (plus or minus) the Laplace operator 𝜕 2 /𝜕𝑥2 +𝜕 2 /𝜕𝑦2
hence the name. Problem 9-1 shows that such an operator can be transformed by a

plus lower-order terms if and only if it is elliptic.

Hodge Theory on Riemannian Manifolds 269

The fundamental fact about elliptic operators on compact manifolds is the fol-
lowing theorem.
Theorem 9.14 (Fredholm Theorem for Elliptic Operators). Suppose 𝑀 is a
compact Riemannian manifold, 𝐸 and 𝐹 are Hermitian vector bundles over 𝑀, and
𝑃 ∶ Γ(𝐸) → Γ(𝐹 ) is an elliptic linear differential operator. Then Ker 𝑃 and Ker 𝑃 ∗
are finite-dimensional, and 𝑃 restricts to a bijection from (Ker 𝑃 )⟂ to (Ker 𝑃 ∗ )⟂ .
Developing the tools to prove this theorem would take us too far afield from
complex manifold theory, but here are some references where proofs can be found:
[Wel08, pp. 136–141], [GH94, pp. 80–100], [Bes87, pp. 456–467], or [War83,

Theorem 9.12 will follow from this once we show that Δ is elliptic.
pp. 220–251].

Lemma 9.15. The Laplace–Beltrami operator Δ on scalar functions is elliptic.

Proof. In any local coordinates (𝑥𝑖 ), the Laplace–Beltrami operator has the coor-
dinate representation
1 𝜕 𝜕𝑢 𝑖𝑗 𝜕 𝑢
2
Δ𝑢 = 𝑖(
𝑔 𝑖𝑗
√det 𝑔 = 𝑔 + terms of order 1
√det 𝑔 𝜕𝑥 𝜕𝑥𝑗 ) 𝜕𝑥𝑖 𝜕𝑥𝑗

(see [LeeRM, Prop. 2.46]). Thus its principal symbol is 𝜎Δ (𝑥, 𝜉) = 𝑔 𝑖𝑗 (𝑥)𝜉𝑖 𝜉𝑗 =
|𝜉|2 , which is invertible whenever 𝜉 ≠ 0. □

Proof of Theorem 9.12. The Fredholm theorem shows that Δ𝑢 = 𝑓 has a solution
if and only if 𝑓 is orthogonal to the kernel of Δ∗ . Now Δ∗ = (−𝑑 ∗ 𝑑)∗ = −𝑑 ∗ 𝑑 =
Δ. Clearly any function that is constant on each connected component of 𝑀 is
in the kernel of Δ. Conversely, if Δ𝑢 = 0, then 0 = (Δ𝑢, 𝑢) = −(𝑑 ∗ 𝑑𝑢, 𝑢) =
−‖𝑑𝑢‖2 , which shows that 𝑢 is constant on each connected component. It follows
that a function is orthogonal to Ker Δ if and only if its integral over each connected
component of 𝑀 is zero. □

Hodge Theory on Riemannian Manifolds

run into a problem: the operator 𝑑 ∗ 𝑑 is not elliptic on forms of degree higher than 0.
When we try to apply the same method to differential forms of higher degree, we

of 𝑑 ∗ 𝑑, so the kernel is infinite-dimensional on manifolds of positive dimension,

One easy way to see that this is true is to note that every closed form is in the kernel

which would contradict the Fredholm theorem if it were elliptic.

forms of all degrees. The Hodge Laplacian is the operator Δ𝑑 ∶ ℰ 𝑞 (𝑀) → ℰ 𝑞 (𝑀)
To get around this problem, we define a new differential operator on differential

Δ𝑑 𝜂 = 𝑑𝑑 ∗ 𝜂 + 𝑑 ∗ 𝑑𝜂,
defined by
270 9. Hodge Theory

where we interpret 𝑑 ∗ to be the zero operator when acting on 0-forms. We define a

harmonic form to be a smooth differential form 𝜂 that satisfies Δ𝑑 𝜂 = 0.
The next proposition describes some basic properties of the Hodge Laplacian.

Proposition 9.16. Let 𝑀 be a compact oriented Riemannian manifold.

(a) Δ𝑑 is formally self-adjoint (equal to its formal adjoint): Δ∗𝑑 = Δ𝑑 .
(b) A differential form 𝛼 ∈ ℰ 𝑞 (𝑀) is harmonic if and only if 𝑑𝛼 = 0 and
𝑑 ∗ 𝛼 = 0.
(c) A 0-form is harmonic if and only if its restriction to each connected com-
ponent of 𝑀 is constant.

Proof. Part (a) follows from the formula for the formal adjoint of a composition

For (b), if 𝑑𝛼 = 0 and 𝑑 ∗ 𝛼 = 0, then it follows immediately from the definition

(Proposition 9.5(b)).

of Δ𝑑 that Δ𝑑 𝛼 = 0. Conversely, suppose Δ𝑑 𝛼 = 0. Then

0 = (Δ𝑑 𝛼, 𝛼) = (𝑑𝑑 ∗ 𝛼, 𝛼) + (𝑑 ∗ 𝑑𝛼, 𝛼) = ‖𝑑 ∗ 𝛼‖2 + ‖𝑑𝛼‖2 ,

which shows that 𝑑𝛼 = 0 and 𝑑 ∗ 𝛼 = 0.

Finally, part (c) follows from the fact that harmonic 0-forms are exactly the
functions that satisfy 𝑑𝑢 = 0, since 𝑑 ∗ is the zero map in this case. □

For a compact oriented Riemannian manifold 𝑀, let ℋ 𝑞 (𝑀) denote the space
of harmonic complex-valued 𝑞-forms, the kernel of Δ𝑑 ∶ ℰ 𝑞 (𝑀) → ℰ 𝑞 (𝑀). By
virtue of the preceding proposition, ℋ 𝑞 (𝑀) is contained in the space 𝒵 𝑞 (𝑀) of
closed 𝑞-forms.
We will show below that Δ𝑑 is elliptic in all degrees. But first, we need some
more facts about principal symbols, which will aid in computations.

Lemma 9.17. Suppose 𝐸 and 𝐹 are smooth Hermitian vector bundles over a Rie-
mannian manifold 𝑀, and 𝑃 ∶ Γ(𝐸) → Γ(𝐹 ) is a linear differential operator of
order 𝑚. Then the principal symbol of its formal adjoint is given by

𝜎𝑃 ∗ (𝑥, 𝜉) = (−1)𝑚 (𝜎𝑃 (𝑥, 𝜉)) ∶ 𝐹𝑥 → 𝐸𝑥 .

∗

Proof. The adjoint 𝑃 ∗ is given by 𝑃 ∗ 𝑣 = ∑𝑖 𝑄𝑖 (𝜑𝑖 𝑣), where (𝜑𝑖 ) is a partition of

unity and 𝑄𝑖 is defined by (9.4). When we ignore all terms in which the argument
𝑣 is differentiated fewer than 𝑚 times, 𝑃 ∗ has the local expression

𝑃 ∗ 𝑣 = (−1)𝑚 𝑝𝛼 𝜕𝑗𝑚 ⋯ 𝜕𝑗1 𝑣𝛾 ℎ𝛽𝛾 𝐻 𝜎𝛼 𝑒𝜎 + lower-order terms.

𝛽𝑗1 …𝑗𝑚
∑∑
𝑗𝑖 𝛼𝛽𝛾𝜎
Hodge Theory on Riemannian Manifolds 271

Now if we assume that the local frames for 𝐸 and 𝐹 are orthonormal (so that ℎ𝛽𝛾
and 𝐻 𝜎𝛼 are identity matrices) and compute the principal symbol of 𝑃 ∗ , we get the
matrix representation

(𝜎𝑃 ∗ )𝛾 (𝑥, 𝜉) = (−1) ∑ 𝑝𝜎 (𝑥)𝜉𝑗𝑞 ⋯ 𝜉𝑗1 ,

𝜎 𝑚 𝛾𝑗1 …𝑗𝑚

𝑗𝑖

which is the transposed conjugate of that of (−1)𝑚 𝜎𝑃 (𝑥, 𝜉). □

Lemma 9.18. Suppose 𝐸, 𝐹 , and 𝐺 are smooth vector bundles over a smooth
manifold 𝑀, and 𝑃 ∶ Γ(𝐸) → Γ(𝐹 ), 𝑄 ∶ Γ(𝐹 ) → Γ(𝐺) are linear differential
operators. Then for all (𝑥, 𝜉) ∈ 𝑇 ∗ 𝑀,
𝜎𝑄∘𝑃 (𝑥, 𝜉) = 𝜎𝑄 (𝑥, 𝜉) ∘ 𝜎𝑃 (𝑥, 𝜉).

► Exercise 9.19. Prove this lemma.

𝑑, 𝑑 ∗ , and Δ𝑑 are given by

Proposition 9.20. On an oriented Riemannian manifold, the principal symbols of

(9.18) 𝜎𝑑 (𝑥, 𝜉)𝛼 = 𝜉 ∧ 𝛼;

(9.19) 𝜎𝑑 ∗ (𝑥, 𝜉)𝛼 = −𝜉 ♯ ⅃ 𝛼;
(9.20) 𝜎Δ𝑑 (𝑥, 𝜉)𝛼 = −|𝜉|2 𝛼.
Thus Δ𝑑 is elliptic.

Proof. To prove (9.18), we use (9.17). Thus given 𝑥 ∈ 𝑀, 𝜉 ∈ 𝑇𝑥∗ 𝑀, and 𝛼 ∈

(Λ𝑥 𝑀)ℂ , let 𝛼̃ be a smooth form on a neighborhood of 𝑥 whose value at 𝑥 is 𝛼, and
𝑞

let 𝑣 be a smooth real-valued function such that 𝑣(𝑥) = 0 and 𝑑𝑣𝑥 = 𝜉. Since 𝑑 is
a first-order differential operator, (9.17) gives
𝜎𝑑 (𝑥, 𝜉)𝛼 = 𝑑(𝑣̃
𝛼 )|𝑥 = (𝑑𝑣 ∧ 𝛼 𝑣 𝑑̃
̃ +✘ ✘𝛼 )|𝑥 = 𝜉 ∧ 𝛼.
✘

covector 𝜉, the map 𝛽 ↦ 𝜉 ♯ ⅃ 𝛽 is the adjoint of 𝛼 ↦ 𝜉 ∧ 𝛼, or in other words, for

Next, for (9.19), we use Lemma 9.17. Thus we need to show that for each real

all 𝛼 ∈ (Λ𝑥 𝑀)ℂ and 𝛽 ∈ (Λ𝑥 𝑀)ℂ ,

𝑞−1 𝑞

(9.21) ⟨𝜉 ∧ 𝛼, 𝛽⟩ = ⟨𝛼, 𝜉 ♯ ⅃ 𝛽⟩.

Choose a local real orthonormal coframe (𝜀1 , … , 𝜀𝑁 ) such that 𝜉 = 𝑎𝜀1 |𝑥 for some
real number 𝑎, and let (𝐸1 , … , 𝐸𝑁 ) be the dual orthonormal frame. Then 𝜉 ♯ = 𝑎𝐸1 .
Both sides of (9.21) are linear over ℂ in 𝛼 and conjugate-linear in 𝛽, so it suffices
to prove it when 𝛼 = 𝜀𝐽 and 𝛽 = 𝜀𝐾 , where 𝐽 and 𝐾 are increasing multi-indices
of lengths 𝑞 − 1 and 𝑞, respectively. There are two cases: If (1, 𝑗1 , … , 𝑗𝑞−1 ) =
(𝑘1 , … , 𝑘𝑞 ), then both sides of (9.21) are equal to 𝑎. If not, then both sides are
equal to zero.
272 9. Hodge Theory

Finally, (9.20) follows from (9.18) and (9.19) together with Lemma 9.18:
(9.22) 𝜎Δ𝑑 (𝑥, 𝜉)𝛼 = −𝜉 ∧ (𝜉 ♯ ⅃ 𝛼) − 𝜉 ♯ ⅃ (𝜉 ∧ 𝛼).
Because interior multiplication by 𝜉 ♯ is an antiderivation, we obtain
𝜉 ♯ ⅃ (𝜉 ∧ 𝛼) = (𝜉 ♯ ⅃ 𝜉) ∧ 𝛼 − 𝜉 ∧ (𝜉 ♯ ⅃ 𝛼) = |𝜉|2 𝛼 − 𝜉 ∧ (𝜉 ♯ ⅃ 𝛼),
where we have used the definition of the sharp operator to obtain 𝜉 ♯ ⅃ 𝜉 = 𝜉(𝜉 ♯ ) =
𝑔(𝜉 ♯ , 𝜉 ♯ ) = 𝑔(𝜉, 𝜉) = |𝜉|2 . Substituting this into (9.22) yields (9.20). □

Here is the main theorem of this section.

Theorem 9.21 (Hodge Theorem for Riemannian Manifolds). Suppose 𝑀 is an
oriented compact Riemannian manifold. Then ℋ 𝑞 (𝑀) is finite-dimensional for
each 𝑞, and the composite map
ℋ 𝑞 (𝑀) ↪ 𝒵 𝑞 (𝑀) → 𝐻dR (𝑀; ℂ)
𝑞
(9.23)
is an isomorphism, where 𝒵 𝑞 (𝑀) is the space of closed 𝑞-forms and the second

on 𝑀 has a unique harmonic representative.

map is projection onto the quotient space. Thus every de Rham cohomology class

Remark. The conclusion of the Hodge theorem holds also for nonoriented Rie-

computing the adjoint of 𝑑 and showing that Δ𝑑 is elliptic. Since complex mani-
mannian manifolds. The proof is very similar, but requires different techniques for

folds are always canonically oriented, it is easier for our purposes to stick with the
oriented case.

Proof. The finite-dimensionality of ℋ 𝑞 (𝑀) follows immediately from the Fred-

Proposition 9.16 showed, in particular, that ℋ 𝑞 (𝑀) is contained in the space

holm theorem.

of closed forms, so the composite map (9.23) makes sense. We just need to show

To see that it is injective, suppose 𝛼 is a harmonic form. Proposition 9.2 shows

it is injective and surjective.

that it is the unique form minimizing the Hodge norm within its cohomology class,

To see that it is surjective, suppose 𝛼 is any closed 𝑞-form on 𝑀. If 𝑞 = 0,

so no other harmonic form represents the same class.

then 𝛼 is already harmonic. If 𝑞 ≥ 1, then 𝑑 ∗ 𝛼 is orthogonal to the kernel of 𝑑 by

Proposition 9.6, and therefore orthogonal to the kernel of Δ𝑑 = Δ∗𝑑 , so the Fredholm
theorem for Δ𝑑 shows that there is a (𝑞 − 1)-form 𝛽 satisfying Δ𝑑 𝛽 = 𝑑 ∗ 𝛼. This
means 𝑑𝑑 ∗ 𝛽 + 𝑑 ∗ 𝑑𝛽 = 𝑑 ∗ 𝛼, or
𝑑𝑑 ∗ 𝛽 = 𝑑 ∗ 𝛼 − 𝑑 ∗ 𝑑𝛽.
The right-hand side of this equation is an element of the image of 𝑑 ∗ , and the left-
hand side is an element of the image of 𝑑 and hence also in the kernel of 𝑑. Since
the image of 𝑑 ∗ is orthogonal to the kernel of 𝑑, the two sides of this equation are
Hodge Theory on Riemannian Manifolds 273

orthogonal to each other and thus must both be zero. That implies 𝑑 ∗ (𝛼 − 𝑑𝛽) = 0.
Since it is also the case that 𝑑(𝛼 − 𝑑𝛽) = 0, it follows that 𝛼 − 𝑑𝛽 is a harmonic
representative of the cohomology class of 𝛼. □

Here is an important application of the Hodge theorem. For any vector spaces 𝑉
and 𝑊 , a bilinear map 𝐵 ∶ 𝑉 × 𝑊 → ℂ is said to be nondegenerate if 𝐵(𝑣, 𝑤) = 0
for all 𝑤 implies 𝑣 = 0, and 𝐵(𝑣, 𝑤) = 0 for all 𝑣 implies 𝑤 = 0.

Corollary 9.22 (Poincaré Duality). Let 𝑀 be a compact oriented smooth 𝑁-

𝐻 𝑘 (𝑀; ℂ) × 𝐻 𝑁−𝑘 (𝑀; ℂ) to ℂ given by

dimensional anifold. There is a well-defined nondegenerate bilinear map from

(𝜂, 𝜁) ↦ 𝜂∧𝜁
∫
(9.24)
𝑀

for any smooth forms 𝜂 and 𝜁 representing their de Rham cohomology classes. Thus
for each 𝑘 there is a complex-linear isomorphism

(9.25) 𝐻 𝑘 (𝑀; ℂ) ≅ 𝐻 𝑁−𝑘 (𝑀; ℂ)∗ ,

and the Betti numbers of 𝑀 satisfy 𝑏𝑘 (𝑀) = 𝑏𝑁−𝑘 (𝑀). Analogous results hold for
cohomology with real coefficients.

Proof. If 𝜂 and 𝜁 are closed and either one is exact, then 𝜂 ∧ 𝜁 is exact, so the
expression ∫𝑀 𝜂 ∧ 𝜁 depends only on the cohomology classes of 𝜂 and 𝜁, and thus

To show it is nondegenerate, choose any Riemannian metric for 𝑀, so that

descends to a well-defined bilinear map on cohomology.

𝐻 (𝑀; ℂ) is isomorphic to ℋ 𝑘 (𝑀; ℂ), which is finite dimensional for each 𝑘.

𝑘

Suppose 𝜂 is a closed 𝑘-form such that ∫𝑀 𝜂 ∧ 𝜁 = 0 for all closed (𝑁 − 𝑘)-forms

𝜁. Without loss of generality, we can take 𝜂 to be the harmonic representative of its
cohomology class. We wish to apply this with 𝜁 = ∗ 𝜂. To see that this is a closed
form, note that 𝜂 is also harmonic because Δ𝑑 𝜂 = Δ𝑑 𝜂 = 0, so 0 = 𝑑 ∗ 𝜂 = ± ∗ 𝑑 ∗ 𝜂,
and because ∗ is an isomorphism, this implies 𝑑 ∗ 𝜂 = 0. Thus 0 = ∫𝑀 𝜂 ∧ (∗ 𝜂) =
‖𝜂‖2 , so 𝜂 = 0. The same argument shows that ∫𝑀 𝜂 ∧𝜁 = 0 for all closed 𝜂 implies
𝜁 = 0, thus proving that the bilinear form is nondegenerate.
The isomorphism in (9.25) is defined by sending [𝜂] ∈ 𝐻 𝑘 (𝑀; ℂ) to the linear
functional [𝜁] ↦ ∫𝑀 𝜂 ∧ 𝜁 in 𝐻 𝑁−𝑘 (𝑀; ℂ)∗ . This is a well-defined complex-linear

we also have an injective linear map from 𝐻 𝑁−𝑘 (𝑀; ℂ) to 𝐻 𝑘 (𝑀; ℂ)∗ given by
map, and the nondegeneracy of (9.24) implies it is injective. On the other hand,

the same formula, which shows that the dimension of 𝐻 𝑁−𝑘 (𝑀; 𝐸) is less than or
equal to that of 𝐻 𝑘 (𝑀; 𝐸 ∗ ). Thus both maps are isomorphisms for dimensional
reasons, and the statement about Betti numbers follows from this. The argument
for real cohomology is essentially identical. □
274 9. Hodge Theory

Hodge Theory on Complex Manifolds

Now suppose 𝑀 is a compact complex manifold of dimension 𝑛 endowed with
a Hermitian metric 𝑔. Then 𝑀 is, in particular, a smooth Riemannian manifold,

𝑁 = 2𝑛 is even, some of the formulas simplify. Here are the main properties of the
so the considerations of the previous section all apply. Since the real dimension

Hodge star operator in this case.

Lemma 9.23. Let 𝑀 be a compact Hermitian 𝑛-manifold.
(a) For 𝛼 ∈ ℰ 𝑝+𝑞 (𝑀), ∗ ∗ 𝛼 = (−1)𝑝+𝑞 𝛼 and 𝑑 ∗ 𝛼 = − ∗ 𝑑 ∗ 𝛼.
(b) ∗ maps ℰ 𝑝,𝑞 (𝑀) isomorphically onto ℰ 𝑛−𝑞,𝑛−𝑝 (𝑀).

that the fact that 𝑇 ′ 𝑀 is orthogonal to 𝑇 ″ 𝑀 under the fiber metric ⟨⋅, ⋅⟩ implies
Proof. Part (a) follows immediately from (9.9) and (9.14). To prove (b), note first

that ℰ 𝑝,𝑞 (𝑀) is orthogonal to ℰ 𝑝 ,𝑞 (𝑀) under the Hodge inner product unless
′ ′

(𝑝, 𝑞) = (𝑝′ , 𝑞 ′ ). For 𝛼 ∈ ℰ 𝑝,𝑞 (𝑀), we will show that ∗ 𝛼 ∈ ℰ 𝑛−𝑞,𝑛−𝑝 (𝑀) by
showing that it is orthogonal to ℰ 𝑟,𝑠 (𝑀) unless (𝑟 + 𝑞, 𝑠 + 𝑝) = (𝑛, 𝑛).
Thus let 𝛽 ∈ ℰ 𝑟,𝑠 (𝑀) be arbitrary, and compute

⟨𝛽, ∗ 𝛼⟩𝑑𝑉𝑔 = 𝛽 ∧ ∗ ∗ 𝛼 = (−1)𝑝+𝑞 𝛽 ∧ 𝛼.

Since 𝛼 is a (𝑞, 𝑝)-form, this last expression is a 2𝑛-form of bidegree (𝑟 + 𝑞, 𝑠 + 𝑝),

which is zero unless (𝑟 + 𝑞, 𝑠 + 𝑝) = (𝑛, 𝑛). □

Example 9.24 (Hodge Star on Riemann Surfaces). If 𝑀 is a Riemann surface

with a Kähler metric 𝑔, for any point 𝑝 ∈ 𝑀 we can choose a holomorphic coordi-
nate 𝑧 = 𝑥+𝑖𝑦 such that (𝜕𝑥 , 𝜕𝑦 ) is an oriented orthonormal basis at 𝑝. It follows that
∗𝑑𝑥 = 𝑑𝑦 and ∗𝑑𝑦 = −𝑑𝑥, so ∗𝑑𝑧 = −𝑖𝑑𝑧 and ∗𝑑𝑧 = 𝑖𝑑𝑧. Since 𝑑𝑧 and 𝑑𝑧 span
the spaces of (1, 0)-forms and (0, 1)-forms, respectively, this implies ∗𝛼 = −𝑖𝛼 for
all (1, 0)-forms and ∗𝛽 = 𝑖𝛽 for all (0, 1)-forms. //
We define two new Laplace-type operators mapping (𝑝, 𝑞)-forms to (𝑝, 𝑞)-forms
on every complex manifold: the Dolbeault Laplacian

Δ𝜕 𝛼 = 𝜕𝜕 ∗ 𝛼 + 𝜕 ∗ 𝜕𝛼,

and its conjugate

Δ𝜕 𝛼 = 𝜕𝜕 ∗ 𝛼 + 𝜕 ∗ 𝜕𝛼.
(We interpret 𝜕 ∗ to be zero on (𝑝, 0)-forms, and 𝜕 ∗ to be zero on (0, 𝑞)-forms.) We
will mostly be concerned with the Dolbeault Laplacian Δ𝜕 because of its relation
with Dolbeault cohomology. The next result is the analogue of Proposition 9.16 for
the Dolbeault Laplacian and its conjugate.
Hodge Theory on Complex Manifolds 275

Proposition 9.25. Let (𝑀, 𝑔) be a compact Hermitian manifold.

(a) Δ𝜕 and Δ𝜕 are formally self-adjoint: Δ𝜕∗ = Δ𝜕 and Δ∗𝜕 = Δ𝜕 .
(b) A differential form 𝛼 is in the kernel of Δ𝜕 if and only 𝜕𝛼 = 0 and 𝜕 ∗ 𝛼 = 0,
and in the kernel of Δ𝜕 if and only 𝜕𝛼 = 0 and 𝜕 ∗ 𝛼 = 0.
(c) A (𝑝, 0)-form is in the kernel of Δ𝜕 if and only if it is holomorphic.

► Exercise 9.26. Prove this proposition.

Our next task is to prove that Δ𝜕 is elliptic, for which we need to compute the
formal adjoint of 𝜕.
Proposition 9.27. For 𝛼 ∈ ℰ 𝑝,𝑞 (𝑀), 𝜕 ∗ 𝛼 = − ∗ 𝜕 ∗ 𝛼.

paid to the types of forms. Let 𝛼 ∈ ℰ 𝑝,𝑞 (𝑀) and 𝛽 ∈ ℰ 𝑝,𝑞−1 (𝑀), and note that
Proof. This is just like the proof of Proposition 9.10, with a little more attention

𝛽 ∧ ∗ 𝛼 is an (𝑛, 𝑛 − 1)-form, so 𝑑(𝛽 ∧ ∗ 𝛼) = 𝜕(𝛽 ∧ ∗ 𝛼). Stokes’s theorem gives

0= 𝑑(𝛽 ∧ ∗ 𝛼)
∫𝑀

= 𝜕(𝛽 ∧ ∗ 𝛼)
∫𝑀

= 𝜕𝛽 ∧ ∗ 𝛼 + (−1)𝑝+𝑞−1 𝛽 ∧ 𝜕∗𝛼
∫𝑀 ∫𝑀

= 𝜕𝛽 ∧ ∗ 𝛼 + 𝛽 ∧ ∗∗𝜕∗𝛼
∫𝑀 ∫𝑀
= (𝜕𝛽, 𝛼) + (𝛽, ∗ 𝜕 ∗ 𝛼). □

Proposition 9.28. The principal symbols of 𝜕, 𝜕 ∗ , and Δ𝜕 are

𝜎𝜕 (𝑥, 𝜉)𝛼 = 𝜉 0,1 ∧ 𝛼,

𝜎𝜕 ∗ (𝑥, 𝜉)𝛼 = −(𝜉 1,0 ) ⅃ 𝛼,
♯

𝜎Δ𝜕 (𝑥, 𝜉)𝛼 = −|𝜉 0,1 | 𝛼 = − 21 |𝜉|2 𝛼,

2

where the real covector 𝜉 is decomposed as 𝜉 = 𝜉 0,1 + 𝜉 1,0 = 𝜉 0,1 + 𝜉 0,1 . Thus Δ𝜕
is elliptic.

Proof. Problem 9-2. □

For a compact Hermitian manifold 𝑀, we define ℋ 𝑝,𝑞 (𝑀) to be the kernel of

Δ𝜕 ∶ ℰ 𝑝,𝑞 (𝑀) → ℰ 𝑝,𝑞 (𝑀). Elements of ℋ 𝑝,𝑞 (𝑀) are called 𝝏-harmonic forms;
by Proposition 9.25, they are the forms that satisfy 𝜕𝛼 = 0 and 𝜕 ∗ 𝛼 = 0. Similarly
276 9. Hodge Theory

𝛼 is 𝝏-harmonic if Δ𝜕 𝛼 = 0. To avoid confusion, sometimes we will refer to

forms that are harmonic in the sense defined earlier (i.e., in the kernel of Δ𝑑 ) as
𝒅-harmonic forms.
The next proposition is proved exactly like its counterpart for 𝑑.
Proposition 9.29. A 𝜕-closed (𝑝, 𝑞)-form 𝛼 is the unique minimizer of the Hodge
norm within its Dolbeault cohomology class if and only if it is 𝜕-harmonic.
The next theorem, like its Riemannian counterpart, was also proved by Hodge
[Hod41]. Its proof is just like that of Theorem 9.21.
Theorem 9.30 (Hodge–Dolbeault Theorem). Let 𝑀 be a compact Hermitian
manifold. Then ℋ 𝑝,𝑞 (𝑀) is finite-dimensional for each 𝑝 and 𝑞, and the composite

ℋ 𝑝,𝑞 (𝑀) ↪ 𝒵 𝑝,𝑞 (𝑀) → 𝐻 𝑝,𝑞 (𝑀)

map

is an isomorphism, where 𝒵 𝑝,𝑞 (𝑀) is the space of 𝜕-closed (𝑝, 𝑞)-forms and the

ogy class on 𝑀 has a unique 𝜕-harmonic representative.

second map is projection onto the quotient space. Thus every Dolbeault cohomol-

► Exercise 9.31. Prove Proposition 9.29 and Theorem 9.30.

For a compact complex manifold 𝑀, recall that the Hodge numbers of 𝑴 are
ℎ𝑝,𝑞 (𝑀) = dim 𝐻 𝑝,𝑞 (𝑀) = dim 𝐻 𝑞 (𝑀; Ω𝑝 ).
The Hodge–Dolbeault theorem shows that ℎ𝑝,𝑞 (𝑀) = dim ℋ 𝑝,𝑞 (𝑀) and therefore
the Hodge numbers are finite for all 𝑝 and 𝑞.

Bundle-Valued Forms

dle-valued forms. Suppose 𝑀 is a complex manifold and 𝐸 → 𝑀 is a holomor-

There is an important generalizaton of the Hodge–Dolbeault theorem to bun-

phic vector bundle. Recall that the Cauchy–Riemann operator 𝜕 𝐸 ∶ ℰ 𝑝,𝑞 (𝑀; 𝐸) →
ℰ 𝑝,𝑞+1 (𝑀; 𝐸) is used to define the Dolbeault cohomology groups 𝐻 𝑝,𝑞 (𝑀; 𝐸) with
coefficients in 𝐸, and the Dolbeault theorem (Thm. 6.19) shows that these are iso-
morphic to the sheaf cohomology groups 𝐻 𝑞 (𝑀; Ω𝑝 (𝐸)).

𝑔 on 𝑀 and a Hermitian fiber metric ⟨⋅, ⋅⟩ on 𝐸. We use these to define a Hermitian

To apply Hodge theory to these groups, we start by choosing a Hermitian metric

fiber metric on the bundle Λ𝑝,𝑞 ⊗ 𝐸 of 𝐸-valued forms by setting

(9.26) ⟨𝛼 ⊗ 𝜎, 𝛽 ⊗ 𝜏⟩ = ⟨𝛼, 𝛽⟩ ⟨𝜎, 𝜏⟩

the second. Here ⟨𝛼, 𝛽⟩ is the pointwise Hodge inner product of the ordinary (𝑝, 𝑞)-
and extending by complex linearity in the first argument and conjugate linearity in

forms 𝛼 and 𝛽, and ⟨𝜎, 𝜏⟩ is the inner product of 𝜎 and 𝜏 defined by the chosen
Hodge Theory on Complex Manifolds 277

Hermitian fiber metric on 𝐸. This defines as usual a global Hermitian inner product
on the space of compactly supported 𝐸-valued forms:

(𝜂1 , 𝜂2 ) = ⟨𝜂1 , 𝜂2 ⟩ 𝑑𝑉𝑔 .

∫
(9.27)
𝑀
Using this inner product, we define the adjoint operator 𝜕 ∗𝐸 and the 𝑬-valued Dol-
beault Laplacian Δ𝜕 𝐸 ∶ ℰ 𝑝,𝑞 (𝑀; 𝐸) → ℰ 𝑝,𝑞 (𝑀; 𝐸) given by
(9.28) Δ𝜕𝐸 = 𝜕 𝐸 𝜕 ∗𝐸 + 𝜕 ∗𝐸 𝜕 𝐸 .
An 𝐸-valued differential form in the kernel of Δ𝜕 𝐸 is said to be 𝝏 𝑬 -harmonic.
Lemma 9.32. If 𝐸 is a holomorphic vector bundle, an 𝐸-valued differential form
𝛼 is 𝜕 𝐸 -harmonic if and only if 𝜕 𝐸 𝛼 = 0 and 𝜕 ∗𝐸 𝛼 = 0.

► Exercise 9.33. Prove this lemma.

Proposition 9.34. If 𝐸 is a holomorphic vector bundle, the operator Δ𝜕𝐸 is elliptic.

Proof. Let (𝑠𝑗 ) be a holomorphic local frame for 𝐸 over an open set 𝑈 ⊆ 𝑀. From
(4.20), in terms of this frame 𝜕 𝐸 has the local expression
𝜕 𝐸 (𝛼 𝑗 ⊗ 𝑠𝑗 ) = (𝜕𝛼 𝑗 ) ⊗ 𝑠𝑗 .
Since the principal symbol of 𝜕 𝐸 is independent of the choice of frame, this shows
that the principal symbol of 𝜕 𝐸 is the same as that of 𝜕 on scalar-valued forms,

𝜎𝜕 𝐸 (𝑥, 𝜉)𝜂 = 𝜉 0,1 ∧ 𝜂.

namely

It then follows from Lemma 9.17 that the principal symbol of 𝜕 ∗𝐸 is the same as that
of 𝜕 ∗ , and therefore
𝜎Δ𝜕 (𝑥, 𝜉)𝜂 = 𝜎Δ𝜕 (𝑥, 𝜉)𝜂 = − 21 |𝜉|2 𝜂,
𝐸

which proves the lemma. □

pose 𝑀 is a compact Hermitian manifold and 𝐸 → 𝑀 is a Hermitian holomorphic

Theorem 9.35 (Hodge–Dolbeault Theorem for Bundle-Valued Forms). Sup-

vector bundle. Then the space ℋ 𝑝,𝑞 (𝑀; 𝐸) of 𝜕 𝐸 -harmonic 𝐸-valued (𝑝, 𝑞)-forms
is finite-dimensional for each 𝑝 and 𝑞, and the composite map
ℋ 𝑝,𝑞 (𝑀; 𝐸) ↪ 𝒵 𝑝,𝑞 (𝑀; 𝐸) → 𝐻 𝑝,𝑞 (𝑀; 𝐸)
is an isomorphism. Thus every 𝐸-valued Dolbeault cohomology class has a unique
𝜕 𝐸 -harmonic representative.

Proof. Just like the proof of Theorem 9.30. □

One application of this theorem is to provide another proof that the space of
holomorphic sections of a holomorphic vector bundle on a compact manifold is
finite-dimensional.
278 9. Hodge Theory

Another Proof of Theorem 3.13. Suppose 𝑀 is a compact manifold and 𝐸 → 𝑀

is a holomorphic vector bundle. Smooth sections of 𝐸 are elements of ℰ 0,0 (𝑀; 𝐸),
and they are holomorphic if and only if they are in the kernel of 𝜕 𝐸 . Because 𝜕 ∗𝐸
acts trivially on (0, 0)-forms, the holomorphic sections are exactly the elements of
ℋ 0,0 (𝑀; 𝐸). Finite-dimensionality then follows from the Hodge–Dolbeault theo-
rem. □

Another application is the following finiteness theorem for sheaf cohomology

groups.
Proposition 9.36 (Finiteness Theorem). Suppose 𝑀 is a compact complex man-
ifold and 𝐸 → 𝑀 is a holomorphic vector bundle. Then the sheaf cohomology
groups 𝐻 𝑞 (𝑀; Ω𝑝 (𝐸)) are all finite-dimensional, and are zero unless 0 ≤ 𝑝, 𝑞 ≤
dim 𝑀.

Proof. The Dolbeault theorem shows that 𝐻 𝑞 (𝑀; Ω𝑝 (𝐸)) ≅ 𝐻 𝑝,𝑞 (𝑀; 𝐸) for each
𝑝 and 𝑞. Once we endow 𝑀 with a Hermitian metric, the Hodge–Dolbeault theorem
then shows that 𝐻 𝑝,𝑞 (𝑀; 𝐸) ≅ ℋ 𝑝,𝑞 (𝑀; 𝐸), which is finite-dimensional. More-
over since there are no nontrivial (𝑝, 𝑞) forms unless 0 ≤ 𝑝, 𝑞 ≤ dim 𝑀, those are the
only values of 𝑝 and 𝑞 for which the groups 𝐻 𝑞 (𝑀; Ω𝑝 (𝐸)) can be nontrivial. □

Using different techniques, one can prove a much more general version of this
finiteness theorem, which says that every coherent analytic sheaf on a compact com-
plex manifold has finite-dimensional cohomology in all degrees. For Stein man-

proved in 1953 [Car53] that if 𝒮 is any coherent analytic sheaf on a Stein manifold
ifolds, there is an even stronger result: the French mathematician Henri Cartan

𝑀, then 𝐻 𝑞 (𝑀; 𝒮 ) = 0 for all 𝑞 > 0. This fact is now known as Cartan’s the-

𝒮 , each stalk 𝒮𝑥 is generated as an 𝒪𝑥 -module by global sections of 𝒮 .) Proofs

orem B. (Theorem A, proved in the same paper, showed that for every such sheaf

of the general finiteness theorem and Cartan’s theorems A and B can be found in
[GR09, pp. 243–246].

Serre Duality
There is an analogue of Poincaré duality for Dolbeault cohomology, and more

dle 𝐸 → 𝑀. Recall the wedge product operation between 𝐸-valued forms and
generally for Dolbeault cohomology with coefficients in a holomorphic vector bun-

𝐸 ∗ -valued forms defined by (4.16). When applied to sections 𝜂 ∈ ℰ 𝑝,𝑞 (𝑀; 𝐸) and
𝜁 ∈ ℰ 𝑛−𝑝,𝑛−𝑞 (𝑀; 𝐸 ∗ ), it yields a scalar-valued (𝑛, 𝑛)-form 𝜂 ∧ 𝜁 , which can be in-
tegrated over 𝑀 (assuming 𝑛 = dimℂ 𝑀). The following theorem was proved in
1955 by Jean-Pierre Serre [Ser55a].
Theorem 9.37 (Serre Duality). Let 𝑀 be a compact complex 𝑛-manifold and let
𝐸 → 𝑀 be a holomorphic vector bundle. There is a well-defined nondegenerate
Hodge Theory on Complex Manifolds 279

bilinear map from 𝐻 𝑝,𝑞 (𝑀; 𝐸) × 𝐻 𝑛−𝑝,𝑛−𝑞 (𝑀; 𝐸 ∗ ) to ℂ given by

(𝜂, 𝜁) ↦ 𝜂∧𝜁
∫
(9.29)
𝑀
for any smooth bundle-valued forms 𝜂 and 𝜁 representing their Dolbeault cohomol-

all 𝑝 and 𝑞:
ogy classes. Consequently, we have the following complex-linear isomorphisms for

(9.30) 𝐻 𝑝,𝑞 (𝑀; 𝐸)∗ ≅ 𝐻 𝑛−𝑝,𝑛−𝑞 (𝑀; 𝐸 ∗ );

(9.31) 𝐻 𝑞 (𝑀; Ω𝑝 (𝐸))∗ ≅ 𝐻 𝑛−𝑞 (𝑀; Ω𝑛−𝑝 (𝐸 ∗ ));
(9.32) 𝐻 𝑞 (𝑀; 𝒪(𝐸))∗ ≅ 𝐻 𝑛−𝑞 (𝑀; 𝒪(𝐾𝑀 ⊗ 𝐸 ∗ )).
In addition, the Hodge numbers of 𝑀 satisfy
(9.33) ℎ𝑝,𝑞 (𝑀) = ℎ𝑛−𝑝,𝑛−𝑞 (𝑀).

𝜂 ∈ ℰ 𝑝,𝑞 (𝑀; 𝐸) is 𝜕 𝐸 -closed and 𝜁 ∈ ℰ 𝑛−𝑝,𝑛−𝑞 (𝑀; 𝐸 ∗ ) is 𝜕 𝐸 ∗ -closed. If 𝜂 = 𝜕 𝐸 𝛾,

Proof. First we show that (9.29) is well defined on cohomology classes. Suppose

then it follows from Proposition 4.16(iii) and Stokes’s theorem that

𝜂∧𝜁 = 𝜕𝐸 𝛾 ∧ 𝜂 = 𝜕(𝛾 ∧ 𝜂) = 𝑑(𝛾 ∧ 𝜂) = 0.

∫𝑀 ∫𝑀 ∫𝑀 ∫𝑀
Similarly, the integral is zero when 𝜁 is 𝜕 𝐸 ∗ -exact.
Choose any Hermitian metric 𝑔 on 𝑀 and a Hermitian fiber metric ⟨⋅, ⋅⟩ℎ on 𝐸,
and let ⟨⋅, ⋅⟩ℎ∗ be the dual metric on 𝐸 ∗ (Problem 7-3). Use these metrics to define
𝜕 𝐸 , 𝜕 ∗𝐸 , Δ𝜕 𝐸 , and ℋ 𝑝,𝑞 (𝑀; 𝐸), and let ̂
ℎ ∶ 𝐸 → 𝐸 ∗ be the conjugate-linear bundle
isomorphism defined in Problem 7-3. We also define a conjugate-linear bundle map
∗𝐸 ∶ ℰ 𝑝,𝑞 (𝑀; 𝐸) → ℰ 𝑛−𝑝,𝑛−𝑞 (𝑀; 𝐸 ∗ )

∗𝐸 (𝛼 ⊗ 𝜎) = ∗𝛼 ⊗ ̂ℎ(𝜎) for 𝛼 ∈ ℰ 𝑝,𝑞 (𝑀), 𝜎 ∈ Γ(𝐸),

by

extended bilinearly. The map ∗𝐸 ∗ ∶ ℰ 𝑝,𝑞 (𝑀; 𝐸 ∗ ) → ℰ 𝑛−𝑝,𝑛−𝑞 (𝑀; 𝐸) is defined

similarly, using the conjugate-linear isomorphism ̂ℎ∗ ∶ 𝐸 ∗ → 𝐸 ∗∗ , which is equal
to ̂
ℎ−1 under the canonical identification 𝐸 ∗∗ ≅ 𝐸. Straightforward computations
using Lemma 9.23 show that
(9.34) ∗𝐸 ∗ ∗𝐸 𝜂 = (−1)𝑝+𝑞 𝜂 for 𝜂 ∈ ℰ 𝑝,𝑞 (𝑀; 𝐸),
(9.35) ∗𝐸 ∗𝐸 ∗ 𝜁 = (−1)𝑝+𝑞 𝜁 for 𝜁 ∈ ℰ 𝑝,𝑞 (𝑀; 𝐸 ∗ ).
We also have an analogue of formula (9.7) for ∗𝐸 acting on ℰ 𝑝,𝑞 (𝑀; 𝐸):
(𝛼 ⊗ 𝜎) ∧ ∗𝐸 (𝛽 ⊗ 𝜏) = (𝛼 ⊗ 𝜎) ∧ (∗𝛽 ⊗ ̂
ℎ(𝜏))
=̂
ℎ(𝜏)(𝜎)𝛼 ∧ ∗𝛽
= ⟨𝜎, 𝜏⟩ℎ ⟨𝛼, 𝛽⟩𝑑𝑉𝑔
= ⟨𝛼 ⊗ 𝜎, 𝛽 ⊗ 𝜏⟩ 𝑑𝑉𝑔 ,
280 9. Hodge Theory

where the expression in the last line is the Hermitian fiber metric on 𝐸-valued forms

We need to compute an explicit expression for the formal adjoint 𝜕 ∗𝐸 . Let

defined by (9.26).

𝜂 ∈ ℰ 𝑝,𝑞−1 (𝑀; 𝐸) and 𝜁 ∈ ℰ 𝑝,𝑞 (𝑀; 𝐸). As in the proof of Proposition 9.27,
we compute

0= 𝑑 (𝜂 ∧ ∗𝐸 𝜁 )
∫𝑀

= 𝜕 (𝜂 ∧ ∗𝐸 𝜁 )
∫𝑀

= 𝜕 𝐸 𝜂 ∧ ∗𝐸 𝜁 + (−1)𝑝+𝑞−1 𝜂 ∧ 𝜕 𝐸 ∗ (∗𝐸 𝜁 )
∫𝑀 ∫𝑀

= 𝜕 𝐸 𝜂 ∧ ∗𝐸 𝜁 + 𝜂 ∧ ∗𝐸 ∗𝐸 ∗ 𝜕 𝐸 ∗ (∗𝐸 𝜁 )
∫𝑀 ∫𝑀
= (𝜕 𝐸 𝜂, 𝜁 ) + (𝜂, ∗𝐸 ∗ 𝜕 𝐸 ∗ ∗𝐸 𝜁 ),

𝜕 ∗𝐸 = −∗𝐸 ∗ 𝜕 𝐸 ∗ ∗𝐸 on ℰ 𝑝,𝑞 (𝑀; 𝐸).

which shows that

To show that (9.29) is nondegenerate, suppose the cohomology class [𝜂] ∈

𝐻 (𝑀; 𝐸) satisfies ∫𝑀 𝜂 ∧ 𝜁 = 0 for all [𝜁] ∈ 𝐻 𝑛−𝑝,𝑛−𝑞 (𝑀; 𝐸 ∗ ). We can take
𝑝,𝑞

𝜂 to be the Δ𝜕𝐸 -harmonic representative of its cohomology class, so that 𝜕 𝐸 𝜂 = 0

and 𝜕 ∗𝐸 𝜂 = 0. Note that ∗𝐸 𝜂 ∈ ℰ 𝑛−𝑝,𝑛−𝑞 (𝑀; 𝐸 ∗ ) satisfies
𝜕 𝐸 ∗ (∗𝐸 𝜂) = ±∗𝐸 ∗𝐸 ∗ 𝜕 𝐸 ∗ (∗𝐸 𝜂) = ±∗𝐸 (−𝜕 ∗𝐸 𝜂) = 0.
Thus ∗𝐸 𝜂 represents a cohomology class in 𝐻 𝑛−𝑝,𝑛−𝑞 (𝑀; 𝐸 ∗ ), and our hypothesis
implies
0= 𝜂 ∧ ∗𝐸 𝜂 = ⟨𝜂, 𝜂⟩ 𝑑𝑉𝑔 = ‖𝜂‖2 ,
∫𝑀 ∫𝑀
so [𝜂] = 0. A nearly identical argument shows that ∫𝑀 𝜂 ∧ 𝜁 = 0 for all 𝜂 implies
[𝜁] = 0.
The isomorphism in (9.30) is defined by sending [𝜁] ∈ 𝐻 𝑛−𝑝,𝑛−𝑞 (𝑀; 𝐸 ∗ ) to the
linear functional [𝜂] ↦ ∫𝑀 𝜂 ∧ 𝜁 in 𝐻 𝑝,𝑞 (𝑀; 𝐸)∗ . This is a well-defined complex-

On the other hand, we also have an injective linear map from 𝐻 𝑝,𝑞 (𝑀; 𝐸) to
linear map, and the Serre duality theorem implies it is injective.

𝐻 𝑛−𝑝,𝑛−𝑞
(𝑀; 𝐸 ∗ )∗ that sends [𝜂] to ([𝜁] ↦ ∫𝑀 𝜂 ∧ 𝜁), which shows that the dimen-
sion of 𝐻 𝑝,𝑞 (𝑀; 𝐸) is less than or equal to that of 𝐻 𝑛−𝑝,𝑛−𝑞 (𝑀; 𝐸 ∗ ). Thus both
maps are isomorphisms for dimensional reasons.

rem, and (9.32) is the 𝑝 = 0 case of (9.30), noting that the canonical bundle 𝐾𝑀 is
The isomorphism (9.31) follows directly from (9.30) and the Dolbeault theo-

equal to Λ𝑛,0 𝑀. Finally, (9.33) follows from (9.30) in the special case in which 𝐸
is the trivial line bundle. □
Hodge Theory on Kähler Manifolds 281

Hodge Theory on Kähler Manifolds

In general on complex manifolds, there is no direct relation between Hodge num-
bers and Betti numbers. But in the special case of a Kähler manifold, it turns out
that there is a very close relationship between them. This should not be surprising,
because the whole idea of Kähler manifolds is that they are the Hermitian mani-
folds in which the relationship between the Riemannian metric structure and the
holomorphic structure is as close as it can be.

pose (𝑀, 𝑔) is a Kähler manifold with Kähler form 𝜔. Define a complex-linear bun-
The key is a set of technical identities that hold on every Kähler manifold. Sup-

dle homomorphism 𝐿𝜔 ∶ ℰ 𝑝,𝑞 (𝑀) → ℰ 𝑝+1,𝑞+1 (𝑀), called the Lefschetz operator,
by
𝐿𝜔 𝜂 = 𝜔 ∧ 𝜂,
and let 𝐿∗𝜔 ∶ ℰ 𝑝+1,𝑞+1 (𝑀) → ℰ 𝑝,𝑞 (𝑀) be its adjoint, so that ⟨𝐿𝜔 𝛼, 𝛽⟩ = ⟨𝛼, 𝐿∗𝜔 𝛽⟩
for all forms 𝛼, 𝛽. The Lefschetz operator is named after Solomon Lefschetz, who
introduced it as part of his statement of the hard Lefschetz theorem (Thm. 9.46 be-

sition, we use square brackets to denote commutators: for two operators 𝑃 and 𝑄,
low). The following identities were first proved by Hodge [Hod41]. In this propo-

the notation [𝑃 , 𝑄] represents 𝑃 ∘ 𝑄 − 𝑄 ∘ 𝑃 .

Proposition 9.38 (The Kähler Identities). On every Kähler manifold, the follow-
ing identities hold:
(a) [𝜕 ∗ , 𝐿𝜔 ] = 𝑖𝜕.
(b) [𝜕 ∗ , 𝐿𝜔 ] = −𝑖𝜕.
(c) [𝐿∗𝜔 , 𝜕] = −𝑖𝜕 ∗ .
(d) [𝐿∗𝜔 , 𝜕] = 𝑖𝜕 ∗ .

follow by taking adjoints, noting that (𝑖𝐴)∗ = −𝑖𝐴∗ and [𝐴, 𝐵]∗ = [𝐵 ∗ , 𝐴∗ ] for any
Proof. We need only prove (a), for then (b) follows by conjugation, and (c) and (d)

operators 𝐴 and 𝐵.
To prove (a), we begin by working on ℂ𝑛 with its Euclidean metric. In standard
holomorphic coordinates, the Kähler metric is 𝑔 = ∑𝑗 𝑑𝑧𝑗 𝑑𝑧𝑗 and its associated
Kähler form is 𝜔 = 2𝑖 ∑𝑗 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑗 (see (8.7)). We will derive a simple expression
for 𝜕 ∗ on Euclidean space.
For a differential form 𝛼 = ∑′𝐽 ,𝐾 𝛼𝐽 𝐾 𝑑𝑧𝐽 ∧ 𝑑𝑧𝐾 , we define 𝜕𝑗 𝛼 to be the form
obtained by applying 𝜕/𝜕𝑧𝑗 to the coefficients of 𝛼 in standard coordinates, and 𝜕𝑗 𝛼
is defined similarly:
′ 𝜕𝛼𝐽 𝐾 ′ 𝜕𝛼𝐽 𝐾
𝜕𝑗 𝛼 = 𝑑𝑧𝐽 ∧ 𝑑𝑧𝐾 , 𝜕𝑗 𝛼 = 𝑑𝑧𝐽 ∧ 𝑑𝑧𝐾 .
∑ 𝜕𝑧𝑗 ∑ 𝜕𝑧𝑗
𝐽 ,𝐾 𝐽 ,𝐾
282 9. Hodge Theory

The following identities are straightforward computations once everything is ex-

panded in standard coordinates, using the fact that the coefficients of the metric are
constants:
𝜕𝛼 = 𝑑𝑧𝑗 ∧ 𝜕𝑗 𝛼,
∑
(9.36)
𝑗

𝜕𝛼 = 𝑑𝑧𝑗 ∧ 𝜕𝑗 𝛼,
∑
(9.37)
𝑗
𝜕𝑗 (𝜕𝑘 ⅃ 𝛼) = 𝜕𝑘 ⅃ (𝜕𝑗 𝛼),
𝜕𝑗 ⟨𝛼, 𝛽⟩ = ⟨𝜕𝑗 𝛼, 𝛽⟩ + ⟨𝛼, 𝜕𝑗 𝛽⟩.
(9.38)
(9.39)

One more identity we will need is

(9.40) ⟨𝑑𝑧𝑗 ∧ 𝛼, 𝛽⟩ = 2⟨𝛼, 𝜕𝑗 ⅃ 𝛽⟩.

We proved in (9.21) that ⟨𝜉 ∧𝛼, 𝛽⟩ = ⟨𝛼, 𝜉 ♯ ⅃ 𝛽⟩ for a real covector 𝜉; if 𝜉 is complex,

it follows by splitting it into its real and imaginary parts that ⟨𝜉 ∧ 𝛼, 𝛽⟩ = ⟨𝛼, 𝜉 ♯ ⅃ 𝛽⟩.
Applying this to 𝜉 = 𝑑𝑧𝑗 and noting that 𝜉 ♯ = 2𝜕𝑗 because the metric coefficients
are 𝑔𝑗𝑘 = 21 𝛿𝑗𝑘 , we obtain (9.40).
Using these identities, we will show that the formal adjoint of 𝜕 on ℂ𝑛 is given
by
𝜕∗𝛽 = − 2𝜕𝑗 ⅃ (𝜕𝑗 𝛽).
∑
(9.41)
𝑗

First note that for any smooth, compactly supported function 𝑓 on ℂ𝑛 ,

𝜕𝑗 𝑓 𝑑𝑉𝑔 = 0,
∫ℂ𝑛

and imaginary parts. Therefore if 𝛼 is a smooth 𝑞-form and 𝛽 is a smooth (𝑞 − 1)-

as can be seen easily by applying the fundamental theorem of calculus to the real

form, both compactly supported,

0= 𝜕𝑗 ⟨𝑑𝑧𝑗 ∧ 𝛼, 𝛽⟩ 𝑑𝑉𝑔
∑∫ℂ𝑛
𝑗

= ⟨𝜕𝑗 (𝑑𝑧𝑗 ∧ 𝛼 ), 𝛽⟩ 𝑑𝑉𝑔 + ⟨𝑑𝑧𝑗 ∧ 𝛼, 𝜕𝑗 𝛽⟩ 𝑑𝑉𝑔

∑∫ℂ𝑛
∑∫ ℂ𝑛
𝑗 𝑗

= ⟨𝑑𝑧𝑗 ∧ (𝜕𝑗 𝛼), 𝛽⟩ 𝑑𝑉𝑔 + ⟨𝛼, 2𝜕𝑗 ⅃ (𝜕𝑗 𝛽)⟩ 𝑑𝑉𝑔

∑∫ℂ𝑛
∑∫ℂ𝑛
𝑗 𝑗

= (𝜕𝛼, 𝛽 ) + (𝛼, 2𝜕𝑗 ⅃ (𝜕𝑗 𝛽)).

∑
𝑗

This proves (9.41).

Hodge Theory on Kähler Manifolds 283

Using these identities together with the fact that interior multiplication is an
antiderivation, we compute

𝜕 ∗ (𝜔 ∧ 𝛽) = − 2𝜕𝑗 ⅃ 𝜕𝑗 ( 2𝑖 𝑑𝑧𝑘 ∧ 𝑑𝑧𝑘 ∧ 𝛽 )

∑ ∑
𝑗 𝑘

= −𝑖 𝜕𝑗 ⅃ ( 𝑑𝑧𝑘 ∧ 𝑑𝑧𝑘 ∧ 𝜕𝑗 𝛽 )
∑
𝑗,𝑘

= −𝑖 )𝑑𝑧
𝑑𝑧𝑘 (𝜕 ✭ ∧ 𝜕𝑗 𝛽 + 𝑖
✭𝑘✭ 𝑑𝑧𝑘 (𝜕𝑗 )𝑑𝑧𝑘 ∧ 𝜕𝑗 𝛽
∑ ✭✭✭𝑗 ∑
✭✭
𝑗,𝑘 𝑗,𝑘

−𝑖 𝑑𝑧 ∧ 𝑑𝑧 ∧ (𝜕𝑗 ⅃ 𝜕𝑗 𝛽 )
𝑘 𝑘
∑
𝑗,𝑘

=𝑖 𝑑𝑧𝑗 ∧ 𝜕𝑗 𝛽 + ( 𝑖
𝑑𝑧𝑘 ∧ 𝑑𝑧𝑘 ) ∧ (− 2𝜕 ⅃ 𝜕 𝛽
∑ ∑2 ∑ 𝑗 𝑗 )
𝑗 𝑘 𝑗
= 𝑖𝜕𝛽 + 𝜔 ∧ (𝜕 𝛽), ∗

which proves (a) in the case of ℂ𝑛 with its Euclidean metric.

Now let (𝑀, 𝑔) be an arbitrary Kähler manifold with Kähler form 𝜔. Since both
operators 𝜕 ∗ and 𝐿𝜔 are defined independently of coordinates, we may compute
[𝜕 ∗ , 𝐿𝜔 ] at a point 𝑥 ∈ 𝑀 in any coordinates we choose. Because 𝑔 is Kähler,

first derivatives at 𝑥 match those of the Euclidean metric (Thm. 8.10(e)). Then the
we can choose holomorphic coordinates such that the metric coefficients and their

above computation shows that [𝜕 ∗ , 𝐿𝜔 ] = 𝑖𝜕 at 𝑥, because 𝜕 ∗ = − ∗ 𝜕 ∗ involves

only one derivative of the metric coefficients, and 𝐿𝜔 involves no differentiation at
all. □

For later use, we also note the following two facts about the Lefschetz operator.
Lemma 9.39. Let (𝑀, 𝑔) be an 𝑛-dimensional Kähler manifold, let 𝜔 be its Käh-
ler form, and let 𝐿𝜔 ∶ ℰ 𝑝,𝑞 (𝑀) → ℰ 𝑝+1,𝑞+1 (𝑀) be the corresponding Lefschetz
operator.
(a) For 𝛼 ∈ ℰ 𝑝,𝑞 (𝑀), [𝐿𝜔 , 𝐿∗𝜔 ]𝛼 = (𝑝 + 𝑞 − 𝑛)𝛼.
(b) For every 0 ≤ 𝑘 ≤ 𝑛, the operator 𝐿𝑘𝜔 = 𝐿𝜔 ∘ ⋯ ∘ 𝐿𝜔 ∶ ℰ 𝑛−𝑘 (𝑀) →
ℰ 𝑛+𝑘 (𝑀) is an isomorphism.

Proof. Given 𝑥0 ∈ 𝑀, Theorem 8.10(d) shows that we can find holomorphic co-
ordinates (𝑧𝑗 ) on a neighborhood of 𝑥0 such that 𝑔𝑗𝑘 (𝑥0 ) = 12 𝛿𝑗𝑘 and thus
𝑖
𝜔 𝑥0 = 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑗 |𝑥 .
2∑𝑗
0

Equation (9.40) shows that the adjoint of the operator 𝛼 ↦ 𝑑𝑧𝑗 ∧ 𝛼 is 𝛽 ↦ 2𝜕𝑗 ⅃ 𝛽.
Using this together with its conjugate, we conclude that the following formula holds
284 9. Hodge Theory

at 𝑥0 :
𝑖
𝐿∗𝜔 𝛼 = − (𝑑𝑧𝑗 ∧)∗ (𝑑𝑧𝑗 ∧)∗ 𝛼 = −2𝑖 𝜕 ⅃ 𝜕 ⅃ 𝛼.
2∑𝑗
∑ 𝑗 𝑗
𝑗

Using the fact that interior multiplication is an antiderivation, we get

𝑖
𝐿∗𝜔 𝐿𝜔 𝛼 = −2𝑖 𝜕𝑗 ⅃ 𝜕𝑗 ⅃ ( 𝑑𝑧𝑘 ∧ 𝑑𝑧𝑘 ∧ 𝛼 )
∑ 2∑
𝑗 𝑘

= 𝜕𝑗 ⅃ (𝛿𝑗𝑘 𝑑𝑧𝑘 ∧ 𝛼 + 𝑑𝑧𝑘 ∧ 𝑑𝑧𝑘 ∧ (𝜕𝑗 ⅃ 𝛼))

∑
𝑗,𝑘

= 𝛿𝑗𝑘 𝛿𝑗𝑘 𝛼 − 𝛿𝑗𝑘 𝑑𝑧𝑘 ∧ (𝜕𝑗 ⅃ 𝛼) − 𝛿𝑗𝑘 𝑑𝑧𝑘 ∧ (𝜕𝑗 ⅃ 𝛼)

∑(
(9.42)
𝑗,𝑘

+ 𝑑𝑧𝑘 ∧ 𝑑𝑧𝑘 ∧ (𝜕𝑗 ⅃ 𝜕𝑗 ⅃ 𝛼))

= 𝑛𝛼 − 𝑑𝑧𝑘 ∧ (𝜕𝑘 ⅃ 𝛼) − 𝑑𝑧𝑗 ∧ (𝜕𝑗 ⅃ 𝛼) + 𝐿𝜔 𝐿∗𝜔 𝛼.

∑ ∑
𝑘 𝑗

To interpret the two intermediate terms, note that for 𝛼 of the form

𝛼 = 𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧𝑗𝑝 ∧ 𝑑𝑧𝑘1 ∧ ⋯ ∧ 𝑑𝑧𝑘𝑞 ,

and for a particular index 𝑗, we have 𝜕𝑗 ⅃ 𝛼 = 0 if 𝑗 is not one of the indices

{𝑗1 , … , 𝑗𝑝 }; while if 𝑗 = 𝑗𝑠 , then

̂
𝑑𝑧𝑗𝑠 ∧ (𝜕𝑗𝑠 ⅃ 𝛼) = (−1)𝑠−1 𝑑𝑧𝑗𝑠 ∧ (𝑑𝑧𝑗1 ∧ ⋯ ∧ 𝑑𝑧 𝑗𝑠 ∧ ⋯ ∧ 𝑑𝑧𝑘𝑞
)
= 𝛼.

Since 𝛼 is a (𝑝, 𝑞)-form, there are 𝑝 indices for which this expression is equal to
𝛼 and the rest are zero, so ∑𝑗 𝑑𝑧𝑗 ∧ (𝜕𝑗 ⅃ 𝛼) = 𝑝𝛼 for 𝛼 of this form; and since
every (𝑝, 𝑞)-form is a linear combination of forms of this type, this holds for all
𝛼 ∈ ℰ 𝑝,𝑞 (𝑀). Similarly ∑𝑘 𝑑𝑧𝑘 ∧ (𝜕𝑘 ⅃ 𝛼) = 𝑞𝛼. Inserting these formulas into

To prove (b), we note that 𝐿𝑘𝜔 is a smooth bundle homomorphism and Λ𝑛−𝑘
ℂ (𝑀)
(9.42) finishes the proof of (a).

and Λℂ (𝑀) are vector bundles with the same rank, so it suffices to show 𝐿𝑘𝜔 is
𝑛+𝑘

injective. Writing 𝑘 = 𝑛 − 𝑗, we will show by induction on 𝑗 that 𝐿𝜔 ∶ Λℂ (𝑀) →

𝑛−𝑗 𝑗

Λℂ (𝑀) is injective.
2𝑛−𝑗

The case 𝑗 = 0 is immediate, because 𝐿𝑛𝜔 is just multiplication by 𝜔𝑛 = 𝑛!𝑑𝑉𝑔 .

Assume that 𝐿𝜔 is injective on (𝑗 − 1)-forms, and suppose 𝛼 is a 𝑗-form such
𝑛−𝑗+1

that 𝐿𝜔 𝛼 = 0. Then 𝐿𝜔 𝛼 = 0, and the fact that interior multiplication is an

𝑛−𝑗 𝑛−𝑗+1
Hodge Theory on Kähler Manifolds 285

antiderivation implies that for every complex vector field 𝑋,

0 = 𝑋 ⅃ 𝐿𝜔 𝛼
𝑛−𝑗+1

= 𝑋 ⅃ (𝜔 ∧ ⋯ ∧ 𝜔 ∧ 𝛼)
= (𝑛 − 𝑗 + 1)(𝑋 ⅃ 𝜔) ∧ 𝐿𝜔 𝛼 + 𝐿𝜔 (𝑋 ⅃ 𝛼)
𝑛−𝑗 𝑛−𝑗+1

= 𝐿𝜔 (𝑋 ⅃ 𝛼),
𝑛−𝑗+1

so the inductive hypothesis shows that 𝑋 ⅃ 𝛼 = 0. Since this is true for every
complex vector field 𝑋, it follows that 𝛼 = 0. □

only case we will need. But the proof actually works when (𝑀, 𝑔) is merely Her-
Remark. We proved the preceding lemma for Kähler manifolds because that is the

mitian and 𝜔 is its fundamental 2-form, because the computations were all carried
out pointwise and did not involve any derivatives of 𝜔.

The next three lemmas and the theorem following them are the main applica-
tions of the Kähler identities.
Lemma 9.40. On every Kähler manifold, the following identities hold:
𝜕 ∗ 𝜕 + 𝜕𝜕 ∗ = 0,
𝜕 ∗ 𝜕 + 𝜕𝜕 ∗ = 0.

Proof. Proposition 9.38 shows that 𝜕 ∗ = −𝑖[𝐿∗𝜔 , 𝜕]. Using this and the fact that
𝜕 ∘ 𝜕 = 0, we compute
𝜕 ∗ 𝜕 + 𝜕𝜕 ∗ = −𝑖[𝐿∗𝜔 , 𝜕]𝜕 − 𝑖𝜕[𝐿∗𝜔 , 𝜕]
= −𝑖𝐿∗𝜔 𝜕𝜕 + 𝑖𝜕𝐿∗𝜔 𝜕 − 𝑖𝜕𝐿∗𝜔 𝜕 + 𝑖𝜕𝜕𝐿∗𝜔 = 0.
The second identity follows from the first by conjugation.
Lemma 9.41. On every Kähler manifold, Δ𝑑 = Δ𝜕 + Δ𝜕 .
□

Proof. Lemma 9.40 yields

Δ𝑑 = (𝜕 + 𝜕)(𝜕 ∗ + 𝜕 ∗ ) + (𝜕 ∗ + 𝜕 ∗ )(𝜕 + 𝜕)
= 𝜕𝜕 ∗ + 𝜕𝜕 ∗ + 𝜕𝜕 ∗ + 𝜕𝜕 ∗ + 𝜕 ∗ 𝜕 + 𝜕 ∗ 𝜕 + 𝜕 ∗ 𝜕 + 𝜕 ∗ 𝜕
+✭
(𝜕𝜕✭∗✭
= (𝜕𝜕 ∗ + 𝜕 ∗ 𝜕) + ✭ 𝜕 ∗✭ (𝜕𝜕✭∗✭
𝜕) + ✭ +✭𝜕 ∗✭
𝜕) + (𝜕𝜕 ∗ + 𝜕 ∗ 𝜕)
✭ ✭

= Δ 𝜕 + Δ𝜕 .
Lemma 9.42. On every Kähler manifold, Δ𝜕 = Δ𝜕 .
□

Proof. From Proposition 9.38 again we get 𝜕 ∗ = 𝑖[𝐿∗𝜔 , 𝜕]. Thus

Δ𝜕 = 𝜕𝜕 ∗ + 𝜕 ∗ 𝜕
= 𝑖𝜕[𝐿∗𝜔 , 𝜕] + 𝑖[𝐿∗𝜔 , 𝜕]𝜕
= 𝑖𝜕𝐿∗𝜔 𝜕 − 𝑖𝜕𝜕𝐿∗𝜔 + 𝑖𝐿∗𝜔 𝜕𝜕 − 𝑖𝜕𝐿∗𝜔 𝜕.
286 9. Hodge Theory

On the other hand, since 𝜕 ∗ = −𝑖[𝐿∗𝜔 , 𝜕],

Δ𝜕 = 𝜕𝜕 ∗ + 𝜕 ∗ 𝜕
= −𝑖𝜕[𝐿∗𝜔 , 𝜕] − 𝑖[𝐿∗𝜔 , 𝜕]𝜕
= −𝑖𝜕𝐿∗𝜔 𝜕 + 𝑖𝜕𝜕𝐿∗𝜔 − 𝑖𝐿∗𝜔 𝜕𝜕 + 𝑖𝜕𝐿∗𝜔 𝜕.

Because 𝜕𝜕 = −𝜕𝜕, these two expressions are equal. □

Theorem 9.43. On every Kähler manifold, Δ𝑑 = 2Δ𝜕 = 2Δ𝜕 .

Proof. Just combine Lemmas 9.41 and 9.42. □

Here is the main theorem in this section.

Theorem 9.44 (Hodge Theorem for Kähler Manifolds). Let 𝑀 be a compact

𝑛-dimensional Kähler manifold.

(a) HODGE DECOMPOSITION: For each 𝑘 = 0, … , 2𝑛, the harmonic forms of

degree 𝑘 have the following direct sum decomposition:

ℋ 𝑘 (𝑀) = ℋ 𝑝,𝑞 (𝑀).

⨁
𝑝+𝑞=𝑘

Therefore, the de Rham groups have a corresponding decomposition

𝐻dR
𝑘
(𝑀; ℂ) ≅ 𝐻 𝑝,𝑞 (𝑀),
⨁
𝑝+𝑞=𝑘

and the Betti numbers and Hodge numbers of 𝑀 are related by

𝑏𝑘 (𝑀) = ℎ𝑘,0 (𝑀) + ℎ𝑘−1,1 (𝑀) + ⋯ + ℎ1,𝑘−1 (𝑀) + ℎ0,𝑘 (𝑀).

(b) HODGE DUALITY: For each 𝑝 and 𝑞, conjugation gives a conjugate-linear

bijection from ℋ 𝑝,𝑞 (𝑀) to ℋ 𝑞,𝑝 (𝑀). Thus ℎ𝑝,𝑞 (𝑀) = ℎ𝑞,𝑝 (𝑀).

Proof. Let 𝛼 be a harmonic 𝑘-form, and for each 𝑝 + 𝑞 = 𝑘 let 𝛼 𝑝,𝑞 be its projec-
tion onto ℰ 𝑝,𝑞 (𝑀). Because Δ𝑑 = 2Δ𝜕 , which maps ℰ 𝑝,𝑞 to itself, it follows that
each component 𝛼 𝑝,𝑞 is 𝜕-harmonic and therefore lies in ℋ 𝑝,𝑞 (𝑀). This shows that
ℋ 𝑘 (𝑀) is the sum of the spaces ℋ 𝑝,𝑞 (𝑀) for 𝑝 + 𝑞 = 𝑘, and since any two such

fact that conjugation maps 𝜕-harmonic (𝑝, 𝑞)-forms to 𝜕-harmonic (𝑞, 𝑝)-forms; but
spaces have trivial intersection, the sum is direct. Hodge duality follows from the

since Δ𝜕 = Δ𝜕 , these are also 𝜕-harmonic. □

Applications of Hodge Theory 287

Applications of Hodge Theory

The Hodge numbers of a compact complex 𝑛-manifold can be arranged in the fol-
lowing array, called the Hodge diamond:

ℎ𝑛,𝑛
ℎ𝑛,𝑛−1 ℎ𝑛−1,𝑛
⋮ ⋱
.

ℎ𝑛,0 ⋯ ⋯ ℎ0,𝑛
..

⋱ ⋮
.

ℎ1,0 ℎ0,1
..

ℎ0,0 .
Serre duality shows that the diamond is symmetric under 180∘ rotation: ℎ𝑝,𝑞 =
ℎ𝑛−𝑝,𝑛−𝑞 . If the manifold is Kähler, then Hodge duality shows that the diamond
is symmetric about its vertical axis (ℎ𝑝,𝑞 = ℎ𝑞,𝑝 ), and combining these two sym-
metries shows that it is also symmetric about its horizontal axis: ℎ𝑝,𝑞 = ℎ𝑛−𝑞,𝑛−𝑝 .
In addition, in the Kähler case, the sum of the Hodge numbers in each horizontal
row equals the corresponding Betti number. This leads to additional topological
obstructions to the existence of a Kähler metric, described in the next corollary.
Corollary 9.45. On a compact Kähler manifold, the odd Betti numbers are even
and the even Betti numbers are positive.

Proof. The statement about the even Betti numbers is Theorem 8.18. If 𝑘 = 2𝑚 + 1
is odd, the Hodge decomposition theorem gives
𝑏𝑘 = (ℎ2𝑚+1,0 + ℎ2𝑚,1 + ⋯ + ℎ𝑚+1,1 ) + (ℎ𝑚,𝑚+1 + ⋯ + ℎ1,2𝑚 + ℎ0,2𝑚+1 ),
and Hodge duality shows that the two sums in parentheses are equal. (If 𝑘 >
dimℂ 𝑀, some of the numbers in this expression are automatically zero, but the
result still holds.) □

We have already seen examples of compact complex manifolds (the Hopf man-
ifolds) that admit no Kähler structure because their second Betti numbers are zero

was discovered in 1976 by William Thurston. If (𝑀, 𝜔) is a symplectic manifold

(Example 8.20). An application of the requirement that odd Betti numbers are even

(a smooth manifold 𝑀 endowed with a symplectic form 𝜔), an almost complex

structure 𝐽 on 𝑀 is said to be compatible with 𝝎 if 𝜔(𝐽 𝑋, 𝐽 𝑌 ) = 𝜔(𝑋, 𝑌 ) and
𝜔(𝑋, 𝐽 𝑋) > 0 for all 𝑋 and 𝑌 . If in addition 𝐽 is integrable, then 𝑔(𝑋, 𝑌 ) =
𝜔(𝑋, 𝐽 𝑌 ) is a Kähler metric. It can be shown that every symplectic manifold admits
a compatible almost complex structure, and it had been conjectured for some time
288 9. Hodge Theory

that perhaps every compact symplectic manifold had a compatible Kähler struc-

1976 by William Thurston: it is a compact symplectic 4-manifold whose first Betti

ture. A counterexample, called the Kodaira–Thurston manifold, was discovered in

number is 3, so it does not have any Kähler structure. See [CdS01, p. 121] for
details.
Here is another important application.
Theorem 9.46 (Hard Lefschetz Theorem). Let 𝑀 be a compact 𝑛-dimensional
Kähler manifold with Kähler form 𝜔, and let 𝐿𝜔 be its Lefschetz operator. Then for
each 𝑘, 𝐿𝑘𝜔 descends to an isomorphism from 𝐻dR𝑛−𝑘
(𝑀; ℂ) to 𝐻dR 𝑛+𝑘
(𝑀; ℂ).

Proof. Because 𝜔 is closed, 𝑑(𝐿𝜔 𝛼) = 𝑑(𝜔 ∧ 𝛼) = 𝜔 ∧ 𝑑𝛼 = 𝐿𝜔 (𝑑𝛼) for every

differential form 𝛼, so 𝐿𝜔 commutes with 𝑑. The same argument shows that 𝐿𝜔
commutes with 𝜕. Using the Kähler identities together with the fact that 𝜕𝜕+𝜕𝜕 = 0,
we can show that 𝐿𝜔 also commutes with the Dolbeault Laplacian:
Δ𝜕 𝐿𝜔 𝛼 = 𝜕𝜕 ∗ 𝐿𝜔 𝛼 + 𝜕 ∗ 𝜕𝐿𝜔 𝛼
= 𝜕(𝐿𝜔 𝜕 ∗ 𝛼 + 𝑖𝜕𝛼) + 𝜕 ∗ 𝐿𝜔 (𝜕𝛼)
= 𝐿𝜔 𝜕𝜕 ∗ 𝛼 + 𝑖𝜕𝜕𝛼 + 𝐿𝜔 𝜕 ∗ 𝜕𝛼 + 𝑖𝜕𝜕𝛼
= 𝐿𝜔 Δ𝜕 𝛼.
Since Δ𝑑 = 2Δ𝜕 , 𝐿𝜔 also commutes with the Hodge Laplacian Δ𝑑 .
The fact that 𝐿𝜔 commutes with 𝑑 implies that it descends to a linear map from
𝐻dR𝑘
(𝑀; ℂ) to 𝐻dR 𝑘+2
(𝑀; ℂ) (still denoted by 𝐿𝜔 ), defined by 𝐿𝜔 ([𝛼]) = [𝜔 ∧ 𝛼].
To show that 𝐿𝜔 ∶ 𝐻dR
𝑘 𝑛−𝑘
(𝑀; ℂ) → 𝐻dR 𝑛+𝑘
(𝑀; ℂ) is injective, suppose 𝐿𝑘𝜔 [𝛼] = 0
for [𝛼] ∈ 𝐻dR𝑛−𝑘
(𝑀). By the Hodge theorem, we can choose 𝛼 to be the harmonic
representative of its cohomology class. Then 𝐿𝑘𝜔 𝛼 = 𝑑𝛽 for some (𝑛 + 𝑘 − 1)-form
𝛽. Because 𝐿𝜔 commutes with Δ𝑑 , we have Δ𝑑 (𝑑𝛽) = Δ𝑑 (𝐿𝑘𝜔 𝛼) = 𝐿𝑘𝜔 (Δ𝑑 𝛼) = 0.
Thus 𝑑𝛽 is harmonic. But all harmonic forms are in the kernel of 𝑑 ∗ , which is
orthogonal to the image of 𝑑, so 𝑑𝛽 is both harmonic and orthogonal to the space
of harmonic forms; thus it is zero. Because 𝐿𝑘𝜔 𝛼 = 0, injectivity of 𝐿𝑘𝜔 on (𝑛 − 𝑘)-
forms (Lemma 9.39(b)) implies 𝛼 = 0.
To show that 𝐿𝑘𝜔 is surjective, let [𝛽] ∈ 𝐻dR
𝑛+𝑘
(𝑀; ℂ) be arbitrary. Again,
we can assume 𝛽 is the harmonic representative of its cohomology class. Lemma
9.39(b) shows that there is some (𝑛 − 𝑘)-form 𝛼 such that 𝐿𝑘𝜔 𝛼 = 𝛽. Then
𝐿𝑘𝜔 (Δ𝑑 𝛼) = Δ𝑑 (𝐿𝑘𝜔 𝛼) = Δ𝑑 𝛽 = 0, and injectivity of 𝐿𝑘𝜔 shows that Δ𝑑 𝛼 = 0.
Thus 𝛼 represents a cohomology class such that 𝐿𝑘𝜔 [𝛼] = [𝛽]. □

This result was claimed by Solomon Lefschetz in 1924 [Lef24], with a proof
that is generally considered to be incorrect, so the theorem was considered “hard”
by Lefschetz’s contemporaries. (According to the Italian-American mathematician
Gian-Carlo Rota, who was an undergraduate at Princeton when Lefschetz was chair
of the Mathematics Department there, it was said that Lefschetz “had never given
Applications of Hodge Theory 289

a completely correct proof, but had never made a wrong guess either” [Rot08, p.
19].) As you can see, with our modern machinery of Hodge theory and Kähler
identities the proof is quite straightforward; but the nickname of the theorem has
stuck anyway.
One immediate consequence of the hard Lefschetz theorem is the following
additional topological constraint on Kähler manifolds.
Corollary 9.47. Let 𝑀 be a compact Kähler 𝑛-manifold. The Lefschetz operator
𝐿𝜔 ∶ 𝐻dR𝑘
(𝑀; ℂ) → 𝐻dR 𝑘+2
(𝑀; ℂ) is injective for 𝑘 ≤ 𝑛 − 1 and surjective for
𝑘 ≥ 𝑛 − 1. Thus the Betti numbers of 𝑀 satisfy 𝑏𝑘 (𝑀) ≤ 𝑏𝑘+2 (𝑀) for 𝑘 ≤ 𝑛 − 1,
and 𝑏𝑘 (𝑀) ≥ 𝑏𝑘+2 (𝑀) for 𝑘 ≥ 𝑛 − 1.
Because the Kähler form is of type (1, 1), it follows that 𝐿𝜔 maps 𝐻 𝑝,𝑞 (𝑀) to
□

𝐻 (𝑀), so the following corollary is also immediate.

𝑝+1,𝑞+1

Corollary 9.48. Let 𝑀 be a compact Kähler 𝑛-manifold and let 𝐿𝜔 be its Lefschetz
operator. For each 𝑝 and 𝑞, the operator 𝐿𝜔
𝑛−𝑝−𝑞

𝐻 𝑝,𝑞 (𝑀) to 𝐻 𝑛−𝑞,𝑛−𝑝 (𝑀). Thus 𝐿𝜔 ∶ 𝐻 𝑝,𝑞 (𝑀) → 𝐻 𝑝+1,𝑞+1 (𝑀) is injective for
restricts to an isomorphism from

𝑝 + 𝑞 ≤ 𝑛 − 1 and surjective for 𝑝 + 𝑞 ≥ 𝑛 − 1; and the Hodge numbers satisfy

ℎ𝑝,𝑞 (𝑀) ≤ ℎ𝑝+1,𝑞+1 (𝑀) for 𝑝 + 𝑞 ≤ 𝑛 − 1 and ℎ𝑝,𝑞 (𝑀) ≥ ℎ𝑝+1,𝑞+1 (𝑀) for 𝑝 + 𝑞 ≥
𝑛 − 1. □
Note that this corollary gives another proof that the Hodge diamond is symmet-
ric about its horizontal axis (which we previously deduced from Serre duality and
Hodge symmetry). In addition, it says that in order for a compact complex manifold
to admit a Kähler metric, its Hodge diamond must have the property that the Hodge
numbers in each vertical column are nondecreasing below the horizontal axis and
nonincreasing above it.
As another application of Hodge theory, we can determine all the Hodge num-
bers of compact Riemann surfaces and complex projective spaces.
Proposition 9.49 (Hodge Numbers of a Riemann Surface). Let 𝑀 be a con-
nected compact Riemann surface of genus 𝑔. The Hodge numbers of 𝑀 are
ℎ0,0 = ℎ1,1 = 1 and ℎ0,1 = ℎ1,0 = 𝑔.

Proof. Example 8.13 showed that 𝑀 admits a Kähler metric. Since 𝑀 is con-
nected and compact, 𝐻dR
0
(𝑀; ℂ) ≅ 𝐻dR
2
(𝑀; ℂ) ≅ ℂ [LeeSM, Prop. 17.6 and
Thm. 17.31]. Thus the Hodge decomposition theorem gives ℎ0,0 = 𝑏0 = 1 and
ℎ1,1 = 𝑏2 = 1.
On the other hand, 𝐻1 (𝑀) is a free abelian group of rank 2𝑔 [LeeTM, Cor.
13.15], so by the universal coefficient theorem, 𝐻dR
1
(𝑀; ℂ) ≅ Hom(𝐻1 (𝑀), ℂ) ≅
ℂ2𝑔 . The statement about ℎ0,1 and ℎ1,0 then follows from the Hodge decomposition
theorem. □
290 9. Hodge Theory

of ℂℙ𝑛 are ℎ𝑝,𝑞 = 0 for 𝑝 ≠ 𝑞, and ℎ𝑝,𝑝 = 1 for 0 ≤ 𝑝 ≤ 𝑛.

Proposition 9.50 (Hodge Numbers of Projective Spaces). The Hodge numbers

Proof. From algebraic topology, 𝐻𝑘 (ℂℙ𝑛 ) ≅ ℤ when 𝑘 is even and 0 ≤ 𝑘 ≤ 2𝑛,

p. 140]). Thus 𝐻dR 𝑘

(ℂℙ𝑛 ; ℂ) ≅ Hom(𝐻𝑘 (ℂℙ𝑛 ), ℂ) is 1-dimensional for 𝑘 even and
and it is zero otherwise (see, for example, [LeeTM, Example 13.35(a)] or [Hat02,

0 ≤ 𝑘 ≤ 2𝑛, and zero otherwise.

If any one of the Hodge numbers ℎ𝑝,𝑞 were nonzero for 𝑝 ≠ 𝑞, then the corre-
sponding Betti number would satisfy 𝑏𝑝+𝑞 ≥ ℎ𝑝,𝑞 + ℎ𝑞,𝑝 ≥ 2, so only ℎ𝑝,𝑝 can be
nonzero. The Hodge decomposition theorem then shows that ℎ𝑝,𝑝 = 𝑏2𝑝 = 1 for
0 ≤ 𝑝 ≤ 𝑛. □

For our next application, recall that smooth complex line bundles on a complex

classes, 𝑐(𝐿) ∈ 𝐻Sing

2
(𝑀; ℤ). In most cases, the classification of holomorphic line
manifold are classified up to smooth isomorphism by their sheaf-theoretic Chern

bundles is different—there may be smooth line bundles that have no holomorphic

structure, or ones that have two or more inequivalent holomorphic structures. The
next proposition says, though, that for projective spaces, the holomorphic classifi-
cation is the same as the smooth classification.

projective space ℂℙ𝑛 , the sheaf-theoretic Chern class map

Proposition 9.51 (Classification of Line Bundles on Projective Space). For each

𝑐 ∶ Pic(ℂℙ𝑛 ) → 𝐻Sing
2
(ℂℙ𝑛 ; ℤ)

is an isomorphism. Thus Pic(ℂℙ𝑛 ) ≅ ℤ and Pic0 (ℂℙ𝑛 ) = {0}. It follows that

every smooth complex line bundle on ℂℙ𝑛 has a unique holomorphic structure up
to isomorphism, and every holomorphic line bundle is isomorphic to 𝐻 𝑑 for some
integer 𝑑.

Proof. The long exact sequence associated with the exponential sheaf sequence
(5.20) contains the following portion:
𝛿∗
𝐻 1 (ℂℙ𝑛 ; 𝒪) → 𝐻 1 (ℂℙ𝑛 ; 𝒪 ∗ ) ⟶ 𝐻 2 (ℂℙ𝑛 ; ℤ) → 𝐻 2 (ℂℙ𝑛 ; 𝒪).

𝐻 0,1 (ℂℙ𝑛 ) and 𝐻 0,2 (ℂℙ𝑛 ), respectively, which are both zero by Proposition 9.50.
By the Dolbeault theorem, the groups on the left and right ends are isomorphic to

Since 𝐻 1 (ℂℙ𝑛 ; 𝒪 ∗ ) is isomorphic to the Picard group of ℂℙ𝑛 and the Chern class
map is 𝑐 = −𝛿∗ , it follows that 𝑐 is an isomorphism from Pic(ℂℙ𝑛 ) to 𝐻 2 (ℂℙ𝑛 ; ℤ) ≅
𝐻Sing
2
(ℂℙ𝑛 ; ℤ).
As noted in the proof of Proposition 9.50, 𝐻2 (ℂℙ𝑛 ) ≅ ℤ and 𝐻1 (ℂℙ𝑛 ) =
0, so it follows from the universal coefficient theorem that 𝐻Sing 2
(ℂℙ𝑛 ; ℤ) ≅
Hom(𝐻2 (ℂℙ𝑛 ), ℤ) ≅ ℤ. Thus Pic(ℂℙ𝑛 ) ≅ ℤ, and since the Chern class map is
injective, its kernel Pic0 (ℂℙ𝑛 ) is trivial. Thus holomorphic line bundles on ℂℙ𝑛
Applications of Hodge Theory 291

are classified by their Chern classes. Since smooth line bundles are also classified
by their Chern classes (Thm. 6.29), it follows that every smooth line bundle has

It remains only to prove that every holomorphic bundle is isomorphic to 𝐻 𝑑 for

exactly one holomorphic structure up to isomorphism.

some 𝑑. In the case of ℂℙ1 , Proposition 6.30 shows that smooth line bundles are
classified by their degrees, and since we now know that the smooth classification

holomorphic line bundle on ℂℙ1 of each degree (up to isomorphism). Since 𝐻 𝑑 has
is the same as the holomorphic classification, it follows that there is exactly one

degree 𝑑 by Example 7.27, this shows that every line bundle on ℂℙ1 is isomorphic
to 𝐻 𝑑 for some 𝑑.
For ℂℙ𝑛 with 𝑛 > 1, we know from the first part of the proof that the Picard
group is isomorphic to ℤ, so there is a holomorphic line bundle 𝐿 → ℂℙ𝑛 that
represents a generator of the group. The hyperplane bundle 𝐻 is thus isomorphic
to 𝐿𝑘 for some integer 𝑘. Problem 3-2 shows that 𝐻 pulls back to the hyperplane
bundle of ℂℙ1 under an embedding 𝐹 ∶ ℂℙ1 → ℂℙ𝑛 ; and 𝐿 pulls back to a bundle
𝐹 ∗ 𝐿 → ℂℙ1 whose 𝑘th tensor power is isomorphic to 𝐹 ∗ 𝐻. Since 𝐹 ∗ 𝐻 generates
Pic(ℂℙ1 ) by the argument in the preceding paragraph, it follows that 𝑘 must be ±1,
which shows that 𝐻 generates Pic(ℂℙ𝑛 ). □

Now we can prove Proposition 2.31, which asserts that every nonsingular pro-
jective algebraic hypersurface is defined by a single homogeneous polynomial. It
is a consequence of the following slightly more general result.
Corollary 9.52. If 𝑆 ⊆ ℂℙ𝑛 is a closed codimension-1 complex submanifold, then
𝑆 is the variety defined by a single homogeneous polynomial 𝑝 ∶ ℂ𝑛+1 → ℂ.

Proof. Given such an 𝑆, Theorem 3.39 shows that there is a holomorphic line bun-
dle 𝐿𝑆 → ℂℙ𝑛 and a holomorphic section 𝜎 ∶ ℂℙ𝑛 → 𝐿𝑆 that vanishes simply
on 𝑆 and nowhere else. By Proposition 9.51, 𝐿𝑆 is isomorphic to 𝐻 𝑑 for some
integer 𝑑, which must be positive because 𝐿𝑆 has a nontrivial holomorphic section
that vanishes somewhere. Theorem 3.36 showed that 𝜎 is the section determined
by a homogeneous polynomial 𝑝 of degree 𝑑 on ℂ𝑛+1 , and thus 𝑆 is the variety
determined by 𝑝. □

submanifold of ℂℙ𝑛 is algebraic. We can now prove it in the special case of hyper-
Recall Chow’s theorem (Thm. 2.29), which says that every closed complex

surfaces.
Corollary 9.53 (Chow’s Theorem for Hypersurfaces). Suppose 𝑆 ⊆ ℂℙ𝑛 is a
closed complex submanifold of codimension 1. Then 𝑆 is algebraic.

Proof. This follows immediately from Corollary 9.52. □

292 9. Hodge Theory

Corollary 9.54 (Projective Hypersurfaces are Connected). Suppose 𝑆 ⊆ ℂℙ𝑛

is a closed complex submanifold of codimension 1. Then 𝑆 is connected.

Proof. Assume for the sake of contradiction that 𝑆 is disconnected, and write 𝑆 =
𝑆1 ∪ 𝑆2 , where 𝑆1 , 𝑆2 are open and closed in 𝑆 and disjoint. Since 𝑆1 and 𝑆2
are closed complex hypersurfaces in ℂℙ𝑛 , Corollary 9.52 shows that each is cut
out by a single homogeneous polynomial. But then Lemma 2.40 implies they must
intersect, which is a contradiction. □

Corollary 9.55 (Holomorphic Forms are Closed and Harmonic). Let 𝜂 be a

smooth (𝑝, 0)-form on a compact Kähler manifold. The following are equivalent:
(a) 𝜂 is holomorphic.
(b) 𝜂 is harmonic.
(c) 𝜂 is closed.

Proof. First assume 𝜂 is holomorphic, that is, 𝜕𝜂 = 0. Proposition 9.25(c) shows

that 𝜂 is 𝜕-harmonic, and therefore also 𝑑-harmonic by Theorem 9.43. Next, if 𝜂
is harmonic, it satisfies 𝑑𝜂 = 0 by Proposition 9.16(b). Finally, if 𝜂 is closed, then
0 = 𝑑𝜂 = 𝜕𝜂 +𝜕𝜂, and taking the (𝑝, 1) part of this equation shows that 𝜂 is 𝜕-closed
and thus holomorphic. □

Example 9.56 (Iwasawa Manifolds are Not Kähler). Let 𝑀 = 𝐺/Γ be an Iwa-
sawa manifold as described in Example 1.20. The holomorphic 1-form 𝑑𝑧3 −𝑧2 𝑑𝑧1
is right-invariant on 𝐺 and therefore preserved by the right action of Γ, so it de-
scends to a holomorphic form on 𝐺/Γ which is not closed. Thus 𝐺/Γ has no Kähler
metric. //
The local 𝜕𝜕-lemma (Cor. 4.15) showed that every closed (𝑝, 𝑞)-form is locally
in the image of 𝑖𝜕𝜕. The next corollary shows that on a compact Kähler manifold,
the same is true globally provided the form is exact.
Corollary 9.57 (Global 𝝏𝝏-Lemma). Suppose 𝑀 is a compact Kähler manifold
and 𝜃 is a 𝑑-exact (𝑝, 𝑞)-form on 𝑀, with 𝑝 and 𝑞 both positive. Then there exists a
smooth (𝑝 − 1, 𝑞 − 1)-form 𝛼 defined on all of 𝑀 such that 𝜃 = 𝑖𝜕𝜕𝛼. If 𝜃 is a real
(𝑝, 𝑝)-form, then 𝛼 can be chosen to be real.

Proof. Because 𝜃 is exact, there is a complex-valued (𝑝 + 𝑞 − 1)-form 𝜂 such that

𝜃 = 𝑑𝜂. As in the proof of the local 𝜕𝜕-lemma (Corollary 4.15), we can assume 𝜂
decomposes as 𝜂 = 𝜂 𝑝,𝑞−1 + 𝜂 𝑝−1,𝑞 . We can also decompose 𝜃 = 𝑑𝜂 into bidegrees
as
𝜃 = 𝜕𝜂 𝑝−1,𝑞 + (𝜕𝜂 𝑝−1,𝑞 + 𝜕𝜂 𝑝,𝑞−1 ) + 𝜕𝜂 𝑝,𝑞−1 ,
Applications of Hodge Theory 293

and the fact that 𝜃 is of type (𝑝, 𝑞) implies that the first and last terms on the right-
hand side are zero. In particular, both 𝜂 𝑝−1,𝑞 and 𝜂 𝑝,𝑞−1 are 𝜕-closed. By the Hodge–
Dolbeault theorem, there are 𝜕-harmonic representatives 𝜎 and 𝜌 for the same Dol-
beault cohomology classes as 𝜂 𝑝−1,𝑞 and 𝜂 𝑝,𝑞−1 , respectively, which means there
exist a (𝑝 − 1, 𝑞 − 1)-form 𝛽 and a (𝑞 − 1, 𝑝 − 1)-form 𝛾 such that

𝜂 𝑝−1,𝑞 = 𝜎 + 𝜕𝛽,
𝜂 𝑝,𝑞−1 = 𝜌 + 𝜕𝛾.

Since 𝜕-harmonic forms on a Kähler manifold are also 𝜕-harmonic, it follows that
𝜎 and 𝜌 are both 𝜕-closed and 𝜕-closed. Therefore, if we set 𝛼 = 𝑖𝛾 − 𝑖𝛽, we have

𝑖𝜕𝜕𝛼 = 𝑖𝜕𝜕(𝑖𝛾) − 𝑖𝜕𝜕(𝑖𝛽)

= 𝜕(𝜕𝛾) + 𝜕(𝜕𝛽)
= 𝜕 (𝜂 𝑝,𝑞−1 − 𝜌) + 𝜕 (𝜂 𝑝−1,𝑞 − 𝜎 )
= 𝜃.

As in the local case, if 𝜃 is a real (𝑝, 𝑝)-form, we can choose 𝜂 to be real and then
take 𝛾 = 𝛽, so that 𝛼 = 𝑖𝛽 − 𝑖𝛽 is real. □

Corollary 9.58. Let 𝑀 be a compact Kähler manifold and 𝐿 → 𝑀 a holomor-

phic line bundle. If 𝜂 is any closed real (1, 1)-form representing the cohomology
class 𝑐1ℝ (𝐿), there is a Hermitian fiber metric on 𝐿 whose Chern connection has
curvature Θ satisfying 2𝜋𝑖 Θ = 𝜂.

Proof. Begin by choosing an arbitrary Hermitian fiber metric ℎ on 𝐿, and let Θ be

the curvature form of its Chern connection. Both 2𝜋𝑖 Θ and 𝜂 are real (1, 1)-forms
representing the first real Chern class of 𝐿, so they differ by an exact form. There-
fore, by the global 𝜕𝜕-lemma, we can find a smooth real function 𝑢 that satisfies

𝑖𝜕𝜕 ( 2𝜋
1
𝑢) = 𝑖
2𝜋
Θ − 𝜂.

Define a new fiber metric by ̃ ̃ be the curvature form of its

ℎ = 𝑒𝑢 ℎ, and let Θ
Chern connection. In terms of any holomorphic local frame 𝑠, (7.20) shows that
̃ can be expressed as
the curvature forms Θ and Θ

Θ = 𝜕𝜕 log |𝑠|2ℎ ,
̃ = 𝜕𝜕 log |𝑠|2 = 𝜕𝜕 log(𝑒𝑢 |𝑠|2 ) = 𝜕𝜕𝑢 + Θ =
Θ 2𝜋
𝜂.
̃ ℎ ℎ 𝑖

Thus ̃
ℎ satisfies the conclusion of the corollary. □
294 9. Hodge Theory

The Lefschetz Theorem on (1, 1)-Classes

fying holomorphic line bundles. Let 𝑀 be a complex manifold and 𝐿 → 𝑀 be a

For our next application of Hodge theory, we return to the question of classi-

holomorphic line bundle, and let 𝑐1ℝ (𝐿) ∈ 𝐻dR𝑘

(𝑀; ℝ) ⊆ 𝐻dR 𝑘
(𝑀; ℂ) be its first

By Theorem 7.14, when 𝑐1ℝ (𝐿) is identified with an element of 𝐻 2 (𝑀; ℂ) under
real Chern class.

the coefficient homomorphism 𝐻 2 (𝑀; ℤ) → 𝐻 2 (𝑀; ℂ). Thus it is integral in the

the de Rham isomorphism, it is the image of the sheaf-theoretic Chern class under

sense defined in Chapter 6. In addition, it is always represented by a form 2𝜋𝑖 Θ of

type (1, 1). If 𝑀 is a compact Kähler manifold, let us say that a cohomology class in
𝐻 𝑘 (𝑀; ℂ) ≅ 𝐻dR 𝑘
(𝑀; ℂ) is of type (𝒑, 𝒒) if it lies in the direct summand 𝐻 𝑝,𝑞 (𝑀)
under the Hodge decomposition.
Theorem 9.59 (Lefschetz Theorem on (1, 1)-Classes). Let 𝑀 be a compact Käh-
ler manifold. Every integral cohomology class of type (1, 1) is the first real Chern
class of a holomorphic line bundle.

Proof. Let [𝜂] ∈ 𝐻dR 2

(𝑀; ℂ) be both integral and of type (1, 1). The hypothesis
that [𝜂] is integral means there is a class 𝛾0 ∈ 𝐻 2 (𝑀; ℤ) that maps to the image
of [𝜂] in 𝐻 2 (𝑀; ℂ) under the coefficient homomorphism. From the long exact
sequence associated to the exponential sheaf sequence, we have
𝑖∗
Pic(𝑀) → 𝐻 2 (𝑀; ℤ) ⟶ 𝐻 2 (𝑀; 𝒪),

where 𝑖 ∶ ℤ ↪ 𝒪 is inclusion. We will show that 𝑖∗ (𝛾0 ) = 0; by exactness of the

sequence above, this implies that 𝛾0 is the sheaf-theoretic Chern class of a holo-
morphic line bundle, and then Theorem 7.14 shows that [𝜂] is its first real Chern
class.
Consider the following commutative diagram of sheaf inclusions:

@ℂ
𝑗
𝑘
ℤ / 𝒪.
𝑖


This yields the following diagram of cohomology groups:

𝐻 2 (𝑀; ℂ) o 𝐻dR
2
(𝑀; ℂ)
𝑗∗ qqqq ℛ
𝑘∗ 𝜋 0,2
8
(9.43)
q

𝐻 2 (𝑀; ℤ) / 𝐻 2 (𝑀; 𝒪) o 𝐻 0,2 (𝑀),

qq
qqq

𝑖∗ 𝒟
 
Applications of Hodge Theory 295

where ℛ and 𝒟 are the isomorphisms given by the de Rham–Weil theorem. The
left-hand triangle commutes by functoriality of sheaf cohomology, and the right-

In our current situation, we are given 𝛾0 ∈ 𝐻 2 (𝑀; ℤ) and [𝜂] ∈ 𝐻dR (𝑀; ℂ)
hand square commutes by Proposition 6.21.
2

such that 𝑗∗ 𝛾0 = ℛ[𝜂] and 𝜂 is of type (1, 1). This implies 𝜋 [𝜂] = 0, and therefore
0,2

𝑖∗ 𝛾0 = 𝑘∗ 𝑗∗ 𝛾0 = 𝑘∗ ℛ[𝜂] = 𝒟 𝜋 0,2 [𝜂] = 0. □

The Riemann–Roch Theorem

As another application of Hodge theory, we will prove two important theorems
regarding holomorphic line bundles on Riemann surfaces. The first is a proof that
the mapping from divisors to the Picard group described in Theorem 3.41 is sur-
jective; the second is the Riemann–Roch theorem, which gives a formula for the
dimension of the space of sections of a holomorphic line bundle.

sheaves of sections. Recall the definition of the Euler characteristic of a sheaf 𝒮

Both proofs make use of the following lemma about Euler characteristics of

of vector spaces: 𝜒(𝒮 ) = ∑𝑘 (−1)𝑘 dim 𝐻 𝑘 (𝑀; 𝒮 ), provided 𝐻 𝑘 (𝑀; 𝒮 ) is finite-

dimensional for all 𝑘 and zero for for all but finitely many values of 𝑘.
Lemma 9.60. Let 𝑀 be a compact Riemann surface and 𝐿 → 𝑀 be a holomorphic
line bundle. For any 𝑝 ∈ 𝑀, let 𝐿{𝑝} denote the point bundle associated with the
hypersurface {𝑝}. Then
(9.44) 𝜒(𝒪(𝐿 ⊗ 𝐿{𝑝} )) = 𝜒(𝒪(𝐿)) + 1,
(9.45) 𝜒(𝒪(𝐿 ⊗ 𝐿∗{𝑝} )) = 𝜒(𝒪(𝐿)) − 1.

Proof. Note that these Euler characteristics are all well defined due to Proposition
9.36. We begin by proving (9.45). Consider the following exact sheaf sequence:
𝑒
(9.46) 0 → ℐ{𝑝} (𝐿) ↪ 𝒪(𝐿) ⟶ (𝐿𝑝 )𝑝 → 0,
where ℐ{𝑝} (𝐿) is the sheaf of holomorphic sections of 𝐿 that vanish at 𝑝; (𝐿𝑝 )𝑝
is the skyscraper sheaf whose stalk at 𝑝 is the fiber 𝐿𝑝 , with all other stalks zero;
and 𝑒 is the sheaf morphism defined be evaluating a section at 𝑝: for any open set
𝑈 ⊆ 𝑀, if 𝑝 ∈ 𝑈 , then 𝑒𝑈 (𝜎) = 𝜎(𝑝), and otherwise 𝑒𝑈 is the zero map.
Proposition 5.16 showed that ℐ{𝑝} (𝐿) ≅ 𝒪(𝐿⊗𝐿∗{𝑝} ), so its Euler characteristic
is defined. Example 6.13(c) showed that (𝐿𝑝 )𝑝 is a fine sheaf, so its cohomology
groups are all zero except for 𝐻 0 (𝑀; (𝐿𝑝 )𝑝 ) ≅ 𝐿𝑝 ; thus 𝜒((𝐿𝑝 )𝑝 ) = dim 𝐿𝑝 = 1.
Proposition 6.10 then implies
(9.47) 𝜒(𝒪(𝐿)) = 𝜒(ℐ{𝑝} (𝐿)) + 𝜒((𝐿𝑝 )𝑝 ) = 𝜒(𝒪(𝐿 ⊗ 𝐿∗{𝑝} )) + 1,

𝐿′ = 𝐿 ⊗ 𝐿{𝑝} and noting that 𝐿′ ⊗ 𝐿∗{𝑝} ≅ 𝐿.

which is equivalent to (9.45). Then (9.44) follows by applying (9.45) to the sheaf
□
296 9. Hodge Theory

Here is our first main result about Riemann surfaces.

Theorem 9.61. Let 𝑀 be a compact Riemann surface. The mapping from divi-
sors to the Picard group of 𝑀 that sends a divisor to the isomorphism class of its
associated line bundle is surjective.

Proof. Let 𝐿 → 𝑀 be a holomorphic line bundle. The Dolbeault theorem shows

that for any nonnegative integer 𝑞, 𝐻 𝑞 (𝑀; 𝒪(𝐿)) ≅ 𝐻 0,𝑞 (𝑀; 𝐿). Since there are no
nontrivial (0, 𝑞) forms on a Riemann surface for 𝑞 ≥ 2, this implies 𝐻 𝑞 (𝑀; 𝒪(𝐿)) =
0 for 𝑞 ≥ 2, and thus 𝜒(𝒪(𝐿)) = dim 𝐻 0 (𝑀; 𝒪(𝐿)) − dim 𝐻 1 (𝑀; 𝒪(𝐿)).
Let 𝑝 ∈ 𝑀 be arbitrary. An easy induction based on Lemma 9.60 shows that
𝜒(𝐿 ⊗ 𝐿𝑘{𝑝} ) = 𝜒(𝐿) + 𝑘 for any positive integer 𝑘. Thus

dim 𝒪(𝑀; 𝐿 ⊗ 𝐿𝑘{𝑝} ) = dim 𝐻 0 (𝑀; 𝐿 ⊗ 𝐿𝑘{𝑝} )

= 𝜒(𝒪(𝐿 ⊗ 𝐿𝑘{𝑝} )) + dim 𝐻 1 (𝑀; 𝐿 ⊗ 𝐿𝑘{𝑝} )
≥ 𝜒(𝒪(𝐿 ⊗ 𝐿𝑘{𝑝} )) = 𝜒(𝒪(𝐿)) + 𝑘,

which is strictly positive for 𝑘 large enough. This implies that 𝐿 ⊗ 𝐿𝑘{𝑝} has a
nontrivial holomorphic section 𝜎. The uniqueness assertion of Theorem 3.41 shows
that 𝐿 ⊗ 𝐿𝑘{𝑝} ≅ 𝐿𝐷1 , where 𝐷1 is the divisor of 𝜎.
If we let 𝐷2 be the divisor 𝑘𝑝, the fact that the mapping from divisors to the
Picard group is a homomorphism implies that 𝐿𝑘{𝑝} ≅ 𝐿𝐷2 and thus (𝐿𝑘{𝑝} )∗ ≅
𝐿−𝐷2 . It follows that

𝐿 ≅ (𝐿 ⊗ 𝐿𝑘{𝑝} ) ⊗ (𝐿𝑘{𝑝} )∗ ≅ 𝐿𝐷1 ⊗ 𝐿−𝐷2 ≅ 𝐿𝐷1 −𝐷2 . □

Corollary 9.62. Every holomorphic line bundle on a connected compact Riemann

surface admits a nontrivial meromorphic section.

Proof. Let 𝐿 → 𝑀 be such a bundle. Theorem 9.61 shows that there is a divisor
𝐷 such that 𝐿 ≅ 𝐿𝐷 , and then Theorem 3.41 shows that there is a meromorphic
section 𝜎 of 𝐿 whose divisor is 𝐷. Since 𝜎 has only finitely many zeros, it is not
trivial. □

𝑀, Div(𝑀) is the group of divisors on 𝑀, Cl(𝑀) (the divisor class group of 𝑀)

Recall the notations from Chapter 3: for a connected compact Riemann surface

is the group of divisors on 𝑀 modulo linear equivalence, and Pic(𝑀) (the Picard
group of 𝑀) is the group of isomorphism classes of holomorphic line bundles on
𝑀. We also define Div0 (𝑀) ⊆ Div(𝑀) to be the subgroup of divisors of degree
zero, and Cl0 (𝑀) ⊆ Cl(𝑀) to be the image of Div0 (𝑀) in Cl(𝑀). As in Chap-
ter 6, Pic0 (𝑀) (the Picard variety of 𝑀) is the subgroup of Pic(𝑀) consisting of
isomorphism classes of line bundles with zero Chern class; for a Riemann surface,
these are exactly the line bundles of degree zero.
Applications of Hodge Theory 297

Corollary 9.63 (The Divisor Class Group and the Picard Group). Suppose 𝑀
is a connected compact Riemann surface. The map from Div(𝑀) to Pic(𝑀) that

an isomorphism from Cl(𝑀) to Pic(𝑀), taking Cl0 (𝑀) to Pic0 (𝑀).

sends a divisor to the isomorphism class of its associated line bundle descends to

Proof. Theorem 3.41 showed that the natural map from Div(𝑀) to Pic(𝑀) de-
scends to an injective homomorphism from Cl(𝑀) to Pic(𝑀), and Theorem 9.61

of its associated line bundle by Theorem 7.22, the image of Cl0 (𝑀) is exactly
shows that it is surjective. Because the degree of a divisor is equal to the degree

Pic0 (𝑀). □

Because of this result, in the algebraic geometry literature, the Picard group is
sometimes defined to be the group Cl(𝑀) of divisors modulo linear equivalence.
The following theorem is fundamental to the study of Riemann surfaces. It

term involving 𝐾 ⊗ 𝐿∗ ); the full strength of the theorem was proved in 1865 by
was first stated in 1857 by Bernhard Riemann [Rie57] as an inequality (without the

Riemann’s student Gustav Roch [Roc65].

Theorem 9.64 (Riemann–Roch). Suppose 𝑀 is a connected compact Riemann
surface of genus 𝑔, and 𝐿 → 𝑀 is a holomorphic line bundle. Then
dim 𝒪(𝑀; 𝐿) = deg 𝐿 + 1 − 𝑔 + dim 𝒪(𝑀; 𝐾 ⊗ 𝐿∗ ).

Proof. As we observed in the proof of Theorem 9.61, for a line bundle 𝐿 on a

compact Riemann surface 𝑀 the Euler characteristic reduces to
𝜒(𝒪(𝐿)) = dim 𝐻 0 (𝑀; 𝒪(𝐿)) − dim 𝐻 1 (𝑀; 𝒪(𝐿)).
Serre duality shows that dim 𝐻 1 (𝑀; 𝒪(𝐿)) = dim 𝐻 0 (𝑀; 𝒪(𝐾 ⊗ 𝐿∗ )); thus
𝜒(𝒪(𝐿)) = dim 𝐻 0 (𝑀; 𝒪(𝐿)) − dim 𝐻 0 (𝑀; 𝒪(𝐾 ⊗ 𝐿∗ )
= dim 𝒪(𝑀; 𝐿) − dim 𝒪(𝑀; 𝐾 ⊗ 𝐿∗ ),
and the Riemann–Roch theorem is equivalent to the claim that
(9.48) 𝜒(𝒪(𝐿)) = deg 𝐿 + 1 − 𝑔
for every holomorphic line bundle 𝐿 → 𝑀.
Note that we already have a formula for dim 𝒪(𝑀; 𝐾). Because 𝐾 ≅ Λ1,0 𝑀,
the holomorphic sections of 𝐾 are exactly the harmonic (1, 0)-forms by Corollary
9.55, and then Proposition 9.49 shows that dim 𝒪(𝑀; 𝐾) = ℎ1,0 (𝑀) = 𝑔.
Theorem 9.61 shows that 𝐿 is isomorphic to the line bundle associated with
some divisor 𝐷. We begin with the case in which 𝐷 = ∑𝑗 𝑛𝑗 𝑝𝑗 is effective (meaning

deg 𝐷 = ∑𝑗 𝑛𝑗 .
that all of its coefficients are nonnegative). We will prove (9.48) by induction on
298 9. Hodge Theory

The base case is deg 𝐷 = 0. For an effective divisor this means 𝐷 = 0, so 𝐿 is

the trivial bundle. In that case, dim 𝒪(𝑀; 𝐿) = 1 (because the only holomorphic
sections of the trivial line bundle are the constants), and dim 𝒪(𝑀; 𝐾 ⊗ 𝐿∗ ) =
dim 𝒪(𝑀; 𝐾) = 𝑔, so (9.48) holds.
Now let 𝑘 ≥ 1, and assume (9.48) holds for every line bundle 𝐿𝐷 in which 𝐷
is effective and has degree less than 𝑘. Let 𝐷 be an effective divisor of degree 𝑘,
and write 𝐷 in the form 𝐷 = 𝐷′ + 𝑝0 , where 𝐷′ has degree 𝑘 − 1 and 𝑝0 is some
point in 𝑀. Then 𝐿𝐷 ≅ 𝐿𝐷′ ⊗ 𝐿{𝑝0 } and deg 𝐿𝐷 = deg 𝐿𝐷′ + 1.
The inductive hypothesis implies 𝜒(𝒪(𝐿𝐷′ )) = deg 𝐿𝐷′ + 1 − 𝑔, and then
Lemma 9.60 gives
𝜒(𝒪(𝐿𝐷 )) = 𝜒(𝒪(𝐿𝐷′ )) + 1 = deg 𝐿𝐷 + 1 − 𝑔.
This completes the proof in the case that 𝐷 is effective.
For a general divisor 𝐷, by separating the terms with positive and negative
coefficients we can write 𝐷 = 𝐷1 − 𝐷2 , where 𝐷1 and 𝐷2 are effective; thus
𝐿𝐷 ≅ 𝐿𝐷1 ⊗ 𝐿∗𝐷 . It follows from Lemma 9.60 and induction that 𝜒(𝒪((𝐿𝐷 )) =
𝜒(𝒪(𝐿𝐷1 )) − deg 𝐿𝐷2 , and therefore the result of the previous paragraph gives
2

𝜒(𝒪((𝐿𝐷 )) = deg 𝐿𝐷1 + 1 − 𝑔 − deg 𝐿𝐷2 = deg 𝐿𝐷 + 1 − 𝑔. □

We will see quite a few applications of the Riemann–Roch theorem in the prob-
lems at the end of this chapter and the next; here is one to get us started.
Corollary 9.65 (Uniqueness of the Holomorphic Structure on ℂℙ1 ). If 𝑀 is a
connected compact Riemann surface of genus 0, then 𝑀 is biholomorphic to ℂℙ1 .

Proof. Suppose 𝑀 has genus 0. Let 𝑝 ∈ 𝑀 be arbitrary, and let 𝐿 denote the
degree-1 bundle 𝐿{𝑝} . Problem 8-15 shows that deg 𝐾𝑀 = 𝑔−2 = −2, so 𝐾𝑀 ⊗𝐿∗
has degree −3 and therefore no nontrivial holomorphic sections. The Riemann–
Roch theorem then shows that dim 𝒪(𝑀; 𝐿) = 2, and Proposition 7.25 shows that
𝑀 is biholomorphic to ℂℙ1 . □

The Picard and Albanese Varieties

If 𝑀 is a complex manifold, recall that the Picard variety Pic0 (𝑀) ⊆ Pic(𝑀)
is defined as the group of isomorphism classes of line bundles on 𝑀 whose sheaf-
theoretic Chern classes are zero. In this section, we will prove that if 𝑀 is a compact
Kähler manifold, then Pic0 (𝑀) has a natural structure as a complex torus. We will
also introduce another complex torus canonically associated with 𝑀, called the
Albanese variety, and prove that these tori are isomorphic when 𝑀 is a connected
compact Riemann surface.
Theorem 9.66. Let 𝑀 be a compact Kähler manifold. The Picard variety Pic0 (𝑀)
has a natural structure as a complex torus whose dimension is equal to ℎ0,1 (𝑀).
Applications of Hodge Theory 299

Proof. The long exact sequence associated with the exponential sheaf sequence
(6.35) reads in part

𝑖∗ 𝜀∗ 𝛿∗
𝐻 1 (𝑀; ℤ) ⟶ 𝐻 1 (𝑀; 𝒪) ⟶ Pic(𝑀) ⟶ 𝐻 2 (𝑀; ℤ).

The subgroup Pic0 (𝑀) is the kernel of 𝛿∗ in the sequence above, and by exactness
it is isomorphic to the quotient group 𝐻 1 (𝑀; 𝒪)/𝑖∗ 𝐻 1 (𝑀; ℤ). By the Dolbeault
theorem, 𝐻 1 (𝑀; 𝒪) is a complex vector space isomorphic to 𝐻 0,1 (𝑀), whose di-
mension is ℎ0,1 (𝑀). So we need only show that 𝑖∗ 𝐻 1 (𝑀; ℤ) is a lattice in this
space.
Consider the following commutative diagram of cohomology groups analogous
to (9.43):

𝐻 1 (𝑀; ℂ) o 𝐻dR
1
(𝑀; ℂ)
𝑗∗ qqqq ℛ
𝑘∗ 𝜋 0,1
8
(9.49)
q

𝐻 1 (𝑀; ℤ) / 𝐻 1 (𝑀; 𝒪) o 𝐻 0,1 (𝑀).

qq
qqq

𝑖∗ 𝒟
 

Since 𝒟 is an isomorphism, it suffices to show that the image of 𝜋 0,1 ∘ ℛ −1 ∘ 𝑗∗ is

a lattice in 𝐻 0,1 (𝑀).
Let 𝑞 = ℎ0,1 (𝑀) = dim 𝐻 0,1 (𝑀). By the Hodge decomposition theorem, the
dimension of 𝐻dR 1
(𝑀; ℂ) is 2𝑞. Lemma 6.27 shows that ℛ −1 𝑗∗ 𝐻 1 (𝑀; ℤ) is a free
abelian subgroup of 𝐻dR1
(𝑀; ℂ) of rank 2𝑞; and we can choose closed real 1-forms
(𝜂 , … , 𝜂 ) whose cohomology classes form a basis for this subgroup over ℤ, and
1 2𝑞

also a basis for 𝐻dR

1
(𝑀; ℂ) over ℂ. By the Hodge theorem, we may choose them
to be harmonic. Then (𝜋 0,1 𝜂 1 , … , 𝜋 0,1 𝜂 2𝑞 ) are 𝜕-harmonic forms whose Dolbeault
cohomology classes generate the image of 𝜋 0,1 ∘ ℛ −1 ∘ 𝑗∗ in 𝐻 0,1 (𝑀). It remains
only to prove that these cohomology classes are linearly independent over ℝ.
Suppose there are real constants 𝑎1 , … , 𝑎2𝑞 such that ∑𝑗=1 𝑎𝑗 𝜋 0,1 𝜂 𝑗 represents
2𝑞

cohomology class is unique, that means the form ∑𝑗=1 𝑎𝑗 𝜋 0,1 𝜂 𝑗 is identically zero,
2𝑞
the zero Dolbeault cohomology class. Because the harmonic representative of a

and by conjugation ∑𝑗=1 𝑎𝑗 𝜋 1,0 𝜂 𝑗 = 0 as well. This implies ∑𝑗=1 𝑎𝑗 𝜂 𝑗 = 0, and

2𝑞 2𝑞

thus all of the 𝑎𝑗 ’s are zero because the 𝜂 𝑗 ’s are linearly independent over ℂ. □

fined as follows. Let Ω1 (𝑀) be the complex vector space of global holomorphic
The other torus canonically associated with a compact Kähler manifold is de-

1-forms on 𝑀, and let Ω1 (𝑀)∗ be its dual space, that is, the space of complex-linear
functions from Ω1 (𝑀) to ℂ. Define a homomorphism 𝜑 ∶ 𝐻1 (𝑀) → Ω1 (𝑀)∗ as
follows: for each singular homology class 𝛾 ∈ 𝐻1 (𝑀), let 𝜑(𝛾) ∈ Ω1 (𝑀)∗ be the
300 9. Hodge Theory

linear functional

𝜑(𝛾)(𝜂) = 𝜂,
∫
(9.50)
𝛾

senting 𝛾. The Albanese variety of 𝑴 denoted by Alb(𝑀), is the following quotient

where the integral is interpreted by integrating over a smooth singular cycle repre-

group:
Alb(𝑀) = Ω1 (𝑀)∗ /𝜑(𝐻1 (𝑀)).

Theorem 9.67. Let 𝑀 be a compact Kähler manifold. The Albanese variety

Alb(𝑀) is a complex torus whose dimension is equal to ℎ0,1 (𝑀).

Proof. Because there are no nontrivial 𝜕-exact (1, 0)-forms, Ω1 (𝑀) = 𝐻 1,0 (𝑀).
Let 𝑞 = dim Ω1 (𝑀) = ℎ1,0 (𝑀) = ℎ0,1 (𝑀). By the Hodge decomposition theorem,
the dimension of 𝐻dR
1
(𝑀; ℂ) is 2𝑞.
Because 𝑀 is compact, 𝐻1 (𝑀) is finitely generated (see [Hat02, Corollaries
A.8 and A.9]). Let 𝑇 ⊆ 𝐻1 (𝑀) denote the torsion subgroup of 𝐻1 (𝑀), so the quo-
tient group 𝐻1 (𝑀)/𝑇 is a finitely generated free abelian group. By the de Rham
and universal coefficient theorems, 𝐻dR
1
(𝑀; ℂ) is isomorphic to Hom(𝐻1 (𝑀), ℂ),
which in turn is isomorphic to Hom(𝐻1 (𝑀)/𝑇 , ℂ). Since 𝐻dR 1
(𝑀; ℂ) has dimen-
sion 2𝑞, it follows that 𝐻1 (𝑀)/𝑇 has rank 2𝑞.
Choose homology classes (𝛾1 , … , 𝛾2𝑞 ) in 𝐻1 (𝑀) that descend to a basis for
the free abelian group 𝐻1 (𝑀)/𝑇 , and choose a basis (𝜂 1 , … , 𝜂 𝑞 ) for the complex
vector space Ω1 (𝑀). Let (𝜀1 , … , 𝜀𝑞 ) be the dual basis for Ω1 (𝑀)∗ , defined by
𝜀𝑗 (𝜂 𝑘 ) = 𝛿𝑗𝑘 . The subgroup 𝜑(𝐻1 (𝑀)) ⊆ Ω1 (𝑀)∗ is generated by the 2𝑞 elements
𝜑(𝛾1 ), … , 𝜑(𝛾2𝑞 ). In terms of the dual basis (𝜀𝑗 ), the linear functional 𝜑(𝛾𝑖 ) has
coordinate representation 𝜑(𝛾𝑖 ) = ∑𝑗 Π𝑖 𝜀𝑗 , where Π𝑖 = 𝜑(𝛾𝑖 )(𝜂 𝑗 ) = ∫𝛾 𝜂 𝑗 .
𝑗 𝑗
𝑖

Let Π be the 𝑞 × 2𝑞 matrix whose entries are Π𝑖 ;

𝑗
concretely, it is

⎛ 𝜂1 … 𝜂1⎞
⎜∫ ∫ ⎟
⎜ 1 ⎟
𝛾 𝛾2𝑞
Π=⎜ ⋮ ⋱ ⋮ ⎟.
⎜ ⎟
⎜∫ 𝜂 … 𝜂𝑞 ⎟
𝑞
⎝ 1
𝛾 ∫𝛾2𝑞 ⎠
This is called the period matrix of 𝑀 with respect to the chosen bases. The columns
of Π, called the periods of 𝑀 with respect to these bases, represent the linear func-
tionals 𝜑(𝛾1 ), … , 𝜑(𝛾2𝑞 ) in terms of the dual basis (𝜀𝑗 ); thus to complete the proof,
we just need to show that these columns are linearly independent over ℝ, so they
generate a lattice in Ω1 (𝑀)∗ , called the period lattice.
Applications of Hodge Theory 301

Corollary 9.55 shows that each holomorphic 1-form 𝜂 𝑗 is harmonic. The

conjugate forms 𝜂 𝑗 are therefore also harmonic, and thus the Hodge decompo-
sition theorem shows that the cohomology classes of (𝜂 1 , … , 𝜂 𝑞 , 𝜂 1 , … , 𝜂 𝑞 ) form
a basis for 𝐻dR1
(𝑀; ℂ) over ℂ. To show that the elements 𝜑(𝛾𝑖 ) are indepen-
dent, suppose ∑𝑖=1 𝑎𝑖 𝜑(𝛾𝑖 ) = 0 for some real constants 𝑎1 , … , 𝑎2𝑞 . This means
2𝑞

∑𝑖 𝑎𝑖 ∫𝛾 𝜂 𝑗 = 0 for each 𝑗, and by conjugation ∑𝑖 𝑎𝑖 ∫𝛾 𝜂 𝑗 = 0 as well. Since

𝐻dR (𝑀; ℂ) ≅ Hom(𝐻1 (𝑀)/𝑇 ; ℂ), this means that 𝛾 = ∑𝑖 𝑎𝑖 𝛾𝑖 is in the kernel of
𝑖 𝑖
1

every homomorphism from 𝐻1 (𝑀)/𝑇 to ℂ; and since 𝐻1 (𝑀)/𝑇 is a free abelian

group, this can only occur if the image of 𝛾 in 𝐻1 (𝑀)/𝑇 is zero. Since the 𝛾𝑖 ’s form
a basis for 𝐻1 (𝑀)/𝑇 , it follows that the coefficients 𝑎𝑖 are all zero. Thus the images
𝜑(𝛾𝑖 ) are ℝ-linearly independent elements of Ω1 (𝑀)∗ , so they span a lattice, which
shows that Alb(𝑀) is a complex 𝑞-torus. □

When we specialize to Riemann surfaces, these two tori are always isomorphic.
The following theorem was essentially proved (in a somewhat different form) by
Niels Henrik Abel and Carl Gustav Jacob Jacobi in the first half of the nineteenth
century. (See also Problem 9-14.)
Theorem 9.68 (Abel–Jacobi). Let 𝑀 be a connected compact Riemann surface.
The complex Lie groups Pic0 (𝑀) and Alb(𝑀) are holomorphically isomorphic,
and both have dimension equal to the genus of 𝑀.

so fix such a metric on 𝑀.

Proof. Example 8.13 showed that every Riemann surface admits Kähler metrics,

Let 𝑞 be the genus of 𝑀, which is equal to ℎ0,1 (𝑀) by Proposition 9.49. If

𝑞 = 0, then both Pic0 (𝑀) and Alb(𝑀) are trivial groups. Assume from now on
that 𝑞 ≥ 1.
The proof of Theorem 9.66 shows that for any real 1-forms (𝜂 1 , … , 𝜂 2𝑞 ) rep-
resenting a basis for the integral classes in 𝐻dR 1
(𝑀; ℂ), Pic0 (𝑀) is isomorphic to
the quotient group 𝐻 0,1 (𝑀)/Λ, where Λ is the lattice generated by the cohomol-
ogy classes ([𝜋 0,1 𝜂 1 ], … , [𝜋 0,1 𝜂 2𝑞 ]). On the other hand, Alb(𝑀) is the quotient
of Ω1 (𝑀)∗ by the lattice Γ generated by the linear functionals 𝜂 ↦ ∫𝛾 𝜂 for some
homology classes (𝛾1 , … , 𝛾2𝑞 ) in 𝐻1 (𝑀) that form a basis for 𝐻1 (𝑀)/𝑇 (which is
𝑖

equal to 𝐻1 (𝑀) itself in this case).

Since Ω1 (𝑀) = 𝐻 1,0 (𝑀), it follows from the 𝑛 = 1 case of Serre duality
that 𝐻 0,1 (𝑀) is isomorphic to Ω1 (𝑀)∗ via the map 𝜎 ∶ 𝐻 0,1 (𝑀) → Ω1 (𝑀)∗ that
sends the cohomology class of a 𝜕-closed (0, 1) form 𝛼 to the linear functional 𝜎([𝛼])
defined by
𝜎([𝛼])(𝜂) =
𝛼 ∧ 𝜂.
∫ 𝑀
We just need to show that 𝜎(Λ) = Γ. To do so, we begin by introducing a more
careful choice of basis for 𝐻1 (𝑀).
302 9. Hodge Theory

𝑎1 𝑏2

𝑏1 𝑏̃1 𝑎2̃ 𝑎2

𝑎1 𝑎1̃ 𝑏̃2 𝑏2
𝑏1 𝑎2

Figure 9.1. A polygon whose quotient is a surface of genus 2

Since 𝑀 is a compact, connected, orientable topological 2-manifold of genus

𝑞, the classification theorem for compact surfaces [LeeTM, Thm. 6.15] shows that
it is homeomorphic to a quotient space of a 4𝑞-sided polygon with edges identified
in pairs according to the scheme 𝑎1 , 𝑏1 , 𝑎−1
1 , 𝑏1 , … , 𝑎𝑞 , 𝑏𝑞 , 𝑎𝑞 , 𝑏𝑞 (see Fig. 9.1
−1 −1 −1

for the case 𝑞 = 2). The images of the 4𝑞 sides in the quotient space (with suitable
parametrizations) are 2𝑞 simple closed curves (thus singular cycles) whose homol-
ogy classes represent a basis for the free abelian group 𝐻1 (𝑀). (See [LeeTM, Thm.

Any two closed curves in 𝑀 that are freely homotopic represent the same ele-
13.14 and Cor. 13.15].)

ment in 𝐻1 (𝑀). One way to see this is to note that a closed curve 𝑎 ∶ [0, 1] → 𝑀
descends to a continuous map 𝑎 ̂ ∶ 𝕊1 → 𝑀, and the homology class represented
by the cycle 𝑎 is equal to 𝑎∗̂ (𝜀), where 𝜀 is a generator of the infinite cyclic group
𝐻1 (𝕊1 ) and 𝑎∗ ∶ 𝐻1 (𝕊1 ) → 𝐻1 (𝑀) is the induced homology homomorphism; then

homology homomorphism [LeeTM, Thm. 13.8]. By replacing the cycles 𝑎𝑗 and 𝑏𝑗

the claim follows from the fact that homotopic continuous maps induce the same

with freely homotopic cycles 𝑎𝑗̃ and 𝑏̃𝑗 as in Fig. 9.1, we can arrange that for each
𝑗, the images of 𝑎𝑗̃ and 𝑏̃𝑗 intersect exactly once, and otherwise the images of all

[LeeSM, Thm. 6.26], we can arrange that the cycles 𝑎𝑗̃ and 𝑏̃𝑗 are smooth cycles
the cycles are pairwise disjoint. Moreover, by the Whitney approximation theorem

topy theorem [LeeSM, Thm. 6.36] we can arrange that the curves 𝑎𝑗̃ and 𝑏̃𝑗 meet
whose images are embedded smooth submanifolds, and by the transversality homo-

as 𝑎𝑗 and 𝑏𝑗 . Choose smooth constant-speed parametrizations 𝑎𝑗 , 𝑏𝑗 ∶ [0, 1] → 𝑀

transversely. From now on, we will drop the tildes and refer to these smooth curves

in such a way that 𝑝𝑗 = 𝑎𝑗 (0) = 𝑏𝑗 (0) = 𝑎𝑗 (1) = 𝑏𝑗 (1) is the point where the curves
meet, and the velocity vectors (𝑎′𝑗 (0), 𝑏′𝑗 (0)) = (𝑎′𝑗 (1), 𝑏′𝑗 (1)) form an oriented basis
for 𝑇𝑝𝑗 𝑀.
Applications of Hodge Theory 303

𝑏𝑗
𝑎𝑗
𝑁
𝐴𝑗

Figure 9.2. The support of 𝛼 𝑗 (the darker shaded region)

For each 𝑗, we will construct closed real 1-forms 𝛼 𝑗 and 𝛽 𝑗 on 𝑀 with the
following properties for all 𝑗, 𝑘 = 1, … , 𝑛:

𝛼𝑘 = 𝛽 𝑘 = 0;
∫ ∫
(9.51)
𝑎𝑗 𝑏𝑗

𝛽𝑘 = − 𝛼 𝑘 = 𝛿𝑗𝑘 ;
∫ ∫
(9.52)
𝑎𝑗 𝑏𝑗

𝜂= 𝛼𝑗 ∧ 𝜂 for every closed 1-form 𝜂;

∫ ∫
(9.53)
𝑎𝑗 𝑀

𝜂= 𝛽𝑗 ∧ 𝜂 for every closed 1-form 𝜂.

∫ ∫
(9.54)
𝑏𝑗 𝑀

First we choose a compact subset 𝐴𝑗 ⊆ 𝑀 diffeomorphic to 𝕊1 ×[0, 1], with the

image of 𝑎𝑗 as one of its boundary components, such that the parametrization of 𝑎𝑗
is consistent with the Stokes orientation of 𝜕𝐴𝑗 . This can be done by letting 𝑁 be
the unit normal vector field along 𝑎𝑗 such that (𝑎′𝑗 (𝑡), 𝑁(𝑡)) is an oriented basis at
each point 𝑎𝑗 (𝑡), and letting 𝐴𝑗 be the set

𝐴𝑗 = {exp𝑎𝑗 (𝑡) (𝑠𝑁(𝑡)) ∶ 𝑡 ∈ [0, 1], 𝑠 ∈ [0, 𝜀]}

for some small 𝜀 > 0 (see Fig. 9.2). By choosing 𝜀 small enough, we can ensure
that 𝐴𝑗 is disjoint from all of the curves 𝑎𝑘 and 𝑏𝑘 for 𝑘 ≠ 𝑗, and such that the curve
𝑏𝑗 enters 𝐴𝑗 on one boundary component exactly once and leaves exactly once on
the other boundary component. (It is an easy consequence of transversality that this
is possible.)
304 9. Hodge Theory

Now let 𝑓 ∶ 𝐴𝑗 → [0, 1] be a smooth function such that 𝑓 ≡ 1 in a neigh-

borhood of the image of 𝑎𝑗 and 𝑓 ≡ 0 in a neighborhood of the other boundary
component of 𝐴𝑗 , and define a smooth closed 1-form 𝛼 𝑗 on 𝑀 by

𝑑𝑓 , on 𝐴𝑗 ,
𝛼𝑗 =
{0, on 𝑀 ∖ supp(𝑑𝑓 ).

This form satisfies ∫𝑎 𝛼 𝑗 = 0 for all 𝑘 because the image of each 𝑎𝑘 is disjoint
from the support of 𝛼 𝑗 . Similarly, ∫𝑏 𝛼 𝑗 = 0 for 𝑘 ≠ 𝑗. For 𝑘 = 𝑗, our choices of
𝑘

orientations guarantee that 𝑏′𝑗 (0) points into 𝐴𝑗 , so there is some 𝑡1 > 0 such that
𝑘

𝑏𝑗 (𝑡) lies in 𝐴𝑗 exactly when 𝑡 ∈ [0, 𝑡1 ], and 𝑏𝑗 (𝑡1 ) is on the boundary component
of 𝐴𝑗 where 𝑓 = 0. Thus

𝛼𝑗 = 𝛼𝑗 = 𝑑𝑓 = 𝑓 (𝑏𝑗 (𝑡1 )) − 𝑓 (𝑏𝑗 (0)) = −1.

∫𝑏𝑗 ∫𝑏𝑗 |[0,𝑡 ∫𝑏𝑗 |[0,𝑡
1] 1]

Now suppose 𝜂 is an arbitrary closed (real or complex) 1-form on 𝑀. Since

𝛼 𝑗 ∧ 𝜂 is supported in 𝐴𝑗 , by Stokes’s theorem we have

𝛼𝑗 ∧ 𝜂 = 𝛼𝑗 ∧ 𝜂 = 𝑑𝑓 ∧ 𝜂 = 𝑑(𝑓 ∧ 𝜂) = 𝑓𝜂 = 𝜂.
∫𝑀 ∫𝐴𝑗 ∫𝐴𝑗 ∫𝐴𝑗 ∫𝜕𝐴𝑗 ∫𝑎𝑗

Finally, we use the same technique to construct an annulus 𝐵𝑗 for each 𝑏𝑗 and
a corresponding 1-form 𝛽 𝑗 satisfying ∫𝑏 𝛽 𝑗 = 0 for all 𝑘 and ∫𝑎 𝛽 𝑗 = 0 for all
𝑘 ≠ 𝑗, and ∫𝑀 𝛽 𝑗 ∧ 𝜂 = ∫𝑏 𝜂 for every closed 1-form 𝜂. In this case our choices
𝑘 𝑘

of orientations ensure that the velocity vector 𝑎′𝑗 (0) = 𝑎′𝑗 (1) is outward-pointing at
𝑗

the point where 𝑎𝑗 and 𝑏𝑗 meet, so there is a time 𝑠1 < 1 such that 𝑎𝑗 (𝑠) ∈ 𝐵𝑗 for
𝑠 ∈ [𝑠1 , 1], and we conclude that ∫𝑎 𝛽 𝑗 = +1. This completes the proof of the
𝑗

It follows from (9.51)–(9.52) that 𝛼 𝑗 and 𝛽 𝑗 yield integer values when integrated
existence of forms satisfying (9.51)–(9.54).

need to check that they generate all such cohomology classes. Since 𝐻0 (𝑀) ≅ ℤ
over smooth singular cycles, so they determine integral cohomology classes. We

is free abelian, the universal coefficient theorem implies that 𝐻Sing 1

(𝑀; ℤ) ≅
Hom(𝐻1 (𝑀), ℤ). It follows from (9.51) and (9.52) that the de Rham cohomol-
ogy classes ([𝛽 1 ], … , [𝛽 𝑛 ], [−𝛼 1 ], … , [−𝛼 𝑛 ]) serve as the dual basis in the group
Hom(𝐻1 (𝑀), ℤ) to the homology basis ([𝑎1 ], … , [𝑎𝑛 ], [𝑏1 ], … , [𝑏𝑛 ]). Thus, con-
sidered as elements of 𝐻dR 1
(𝑀; ℂ), they span the subgroup of integral cohomology
classes, and as noted above, their (0, 1)-parts span the lattice Λ.
Let 𝜂 ∈ Ω1 (𝑀) be arbitrary. Corollary 9.55 shows that 𝜂 is closed. It then
follows from (9.53) that when 𝜎([(𝛼 𝑖 )0,1 ]) is applied to 𝜂, the result is

𝜎([(𝛼 𝑖 )0,1 ])(𝜂) = (𝛼 𝑖 )0,1 ∧ 𝜂 = 𝛼𝑖 ∧ 𝜂 = 𝜂.

∫𝑀 ∫𝑀 ∫𝑎𝑖
Applications of Hodge Theory 305

Similarly, 𝜎([(𝛽 𝑖 )0,1 ])(𝜂) = ∫𝑏 𝜂. Thus the image of Λ under 𝜎 is exactly the lattice
Γ.
𝑖
□

In the case of a connected compact Riemann surface 𝑀, the torus Alb(𝑀) is

usually called the Jacobian variety (or sometimes just the Jacobian) of 𝑀, denoted

instead to be the isomorphic torus Pic0 (𝑀).)

by Jac(𝑀). (Because of the preceding theorem, some authors define the Jacobian

Historically, Jacobian varieties of Riemann surfaces were constructed first; the

Albanese and Picard varieties came later as generalizations to higher-dimensional
Kähler manifolds. In higher dimensions, the Albanese and Picard varieties may
not be isomorphic, but they are dual to each other in a sense described in Problem

projective algebraic varieties when 𝑀 is projective; see Problem 10-12 for a proof
9-13. (The reason for the word “variety” in the names is because they are always

in the Riemann surface case.)

Here is an important application of the Jacobian variety of a curve.
Theorem 9.69. Every connected compact Riemann surface of genus 1 is biholo-
morphic to its Jacobian variety. Therefore, every holomorphic structure on 𝕊1 × 𝕊1
is biholomorphic to a quotient ℂ/Λ for some lattice Λ ⊆ ℂ.

Proof. Let 𝑀 be a connected compact Riemann surface of genus 1. Since the

space Ω1 (𝑀) = 𝐻 1,0 (𝑀) is 1-dimensional, we can choose a holomorphic section
𝜁 spanning Ω1 (𝑀). Note that Λ1,0 𝑀 is equal to the canonical bundle 𝐾, which has
degree 0 in this case by the result of Problem 8-15. It follows that 𝜁 cannot have
any zeros, because if it did, the degree of 𝐾 would be equal to the number of zeros
counted with multiplicity by Theorem 7.22.

ℂ/Γ, where Γ ⊆ ℂ is the lattice spanned by the numbers ( ∫𝑎 𝜁, ∫𝑏 𝜁 ) for some

As in the proof of Proposition 9.67, we will represent Jac(𝑀) as the quotient

piecewise-smooth loops 𝑎, 𝑏 representing a basis for 𝐻1 (𝑀).

Choose a point 𝑝0 ∈ 𝑀, and define a map Φ ∶ 𝑀 → Jac(𝑀) as follows: for
each 𝑞 ∈ 𝑀, Φ(𝑞) is the image in ℂ/Γ of ∫𝑝 𝜁 , where the integral is interpreted
𝑞

as the line integral ∫𝛾 𝜁 for some piecewise smooth curve segment 𝛾 from 𝑝0 to 𝑞.
0

First we note that Φ is well defined. If 𝛾1 and 𝛾2 are two such curves, then the path
product 𝛾1 ⋅ 𝛾2−1 is a loop and therefore represents a homology class in 𝐻1 (𝑀);
thus ∫𝛾 𝜁 and ∫𝛾 𝜁 differ by an element of the lattice Γ and their projections onto
1 2

To see that Φ is holomorphic, let 𝑞0 ∈ 𝑀 be arbitrary and choose a holomor-

Jac(𝑀) are equal.

phic coordinate 𝑧 on a disk 𝑈 ⊆ 𝑀 centered at 𝑞0 . Because 𝜁 is a closed (1, 0)-form,

there is some holomorphic function ℎ defined on 𝑈 such that 𝜁|𝑈 = 𝑑ℎ. Because
the quotient map ℂ → Jac(𝑀) is a holomorphic covering map, we can use the stan-
dard coordinate in ℂ as a local holomorphic coordinate on Jac(𝑀); let Φ ̂ denote the
coordinate representation of Φ with respect to these coordinates. For 𝑧 ∈ 𝑈 , the
306 9. Hodge Theory

fundamental theorem of calculus gives

̂
Φ(𝑧) ̂ 0) +
= Φ(𝑞 ̂ 0 ) + ℎ(𝑧) − ℎ(𝑞0 ),
𝜁 = Φ(𝑞
∫𝛾𝑧
where 𝛾𝑧 is, say, the radial path in 𝑈 from 𝑞0 to 𝑧. This shows that Φ is holomorphic
in 𝑈 . Moreover, since the coordinate representation of Φ in this chart is given by a
line integral of the closed form 𝜁 , its differential there is equal to 𝜁; so the fact that
𝜁 never vanishes implies that Φ is a holomorphic immersion.
Next we show that Φ is injective. Suppose 𝑝, 𝑞 are distinct points in 𝑀 such
that Φ(𝑝) = Φ(𝑞). Letting 𝛾1 and 𝛾2 be piecewise smooth paths from 𝑝0 to 𝑝 and 𝑞,
respectively, this means ∫𝛾 𝜁 differs from ∫𝛾 𝜁 by an element of Γ, or equivalently
that ∫𝛾 −1 ⋅𝛾 𝜁 ∈ Γ. Since every element of Γ is the integral of 𝜁 over some piecewise
1 2

smooth closed curve in 𝑀, by setting 𝛾 = 𝛾1−1 ⋅ 𝛾2 ⋅ 𝜎 for some loop 𝜎 based at 𝑞,

1 2

we can arrange that ∫𝛾 𝜁 = 0 for a path 𝛾 from 𝑝 to 𝑞.

there is a global meromorphic function 𝑓 on 𝑀 whose divisor is equal to 𝑝 − 𝑞. It

The key fact is Lemma 9.70 below, which shows that under this hypothesis,

then follows from Proposition 7.25 that 𝑀 is biholomorphic to ℂℙ1 . This contra-
dicts the assumption that 𝑀 has genus 1, and completes the proof that Φ is injec-
tive. Since Φ is an injective holomorphic immersion between connected compact
manifolds of the same dimension, it is an open and closed map and therefore also
surjective. Thus it is a biholomorphism, and the theorem is proved. □

Here is the lemma that was used in the preceding proof.

Lemma 9.70. Let 𝑀 be a compact Riemann surface, and 𝑝, 𝑞 ∈ 𝑀 be distinct
points. Suppose there is a piecewise-smooth curve 𝛾 from 𝑝 to 𝑞 such that ∫𝛾 𝜁 =
0 for every holomorphic (1, 0)-form 𝜁 . Then there exists a global meromorphic
function on 𝑀 whose divisor is equal to 𝑝 − 𝑞.

Proof. We first consider the case in which the image of 𝛾 ∶ [0, 1] → 𝑀 is con-
tained in a holomorphic coordinate chart 𝑈 biholomorphic to the unit disk. Let 𝑧
be the coordinate function in 𝑈 , and let 𝑎, 𝑏 be the coordinates of the points 𝑝 and
𝑞, respectively. Let 𝜓 ∈ 𝐶 ∞ (𝑀; [0, 1]) be a smooth bump function that is sup-
ported in 𝑈 and equal to 1 on a smaller disk 𝐷𝑟 = {𝑧 ∈ 𝑈 ∶ |𝑧| < 𝑟} for some
1 > 𝑟 > max(|𝑎|, |𝑏|); and define a smooth function 𝑢 ∶ 𝑀 ∖ {𝑝, 𝑞} → ℂ by setting
𝑧−𝑎
𝑢(𝑧) = 𝜓(𝑧) + 1 − 𝜓(𝑧)
𝑧−𝑏
for 𝑧 ∈ 𝑈 ∖ {𝑎, 𝑏} and extending it to be equal to 1 outside the support of 𝜓. Then
𝑢 does not vanish anywhere on 𝑀 ∖ {𝑝, 𝑞}, because it is equal to 1 away from the
support of 𝜓, is nonvanishing by definition in 𝐷𝑟 (0) ∖ {𝑎, 𝑏}, and if 𝑢(𝑧) = 0 for
some 𝑟 ≤ |𝑧| ≤ 1, we would have 𝑧 = 𝜓(𝑧)𝑎 + (1 − 𝜓(𝑧))𝑏 by direct computation,
meaning that 𝑧 is on the line segment connecting 𝑎 and 𝑏, contradicting our choice
of 𝑟. Thus the (0, 1)-form 𝛼 = 𝜕𝑢/𝑢 is smooth everywhere outside of 𝐷𝑟 (0) and
Applications of Hodge Theory 307

vanishes identically on 𝐷𝑟 (0) ∖ {𝑎, 𝑏}, so it extends to a smooth, 𝜕-closed (0, 1)-
form on all of 𝑀.
Choose some Kähler metric on 𝑀. We will show that 𝛼 is orthogonal to the
space of 𝜕-harmonic (0, 1)-forms. Because each holomorphic (1, 0)-form 𝜁 is 𝜕-
harmonic and conjugation takes ℋ 1,0 (𝑀) to ℋ 0,1 (𝑀), the 𝜕-harmonic (0, 1)-forms
are exactly those of the form 𝜁 for 𝜁 ∈ Ω1 (𝑀). Thus we have to show (𝛼, 𝜁) =
∫𝑀 𝛼 ∧ ∗𝜁 = 0 for all such 𝜁. Since ∗𝜁 = −𝑖𝜁 by the result of Example 9.24, this
is equivalent to ∫𝑀 𝛼 ∧ 𝜁 = 0.
For small 𝜀 > 0, let 𝐷𝜀 (𝑎), 𝐷𝜀 (𝑏) ⊆ 𝑈 be the coordinate disks of radius 𝜀 cen-
tered at 𝑎 and 𝑏, respectively, and let 𝑈𝜀 = 𝑈 ∖(𝐷𝜀 (𝑎)∪𝐷𝜀 (𝑏)). For any holomorphic
1-form 𝜁 , using the fact that 𝛼 is supported in 𝑈 , we compute

𝜕𝑢 𝑑𝑢
𝛼 ∧ 𝜁 = lim 𝛼 ∧ 𝜁 = lim ∧ 𝜁 = lim ∧ 𝜁.
∫𝑀 𝜀→0 ∫𝑈
𝜀
𝜀→0 ∫𝑈 𝜀
𝑢 𝜀→0 ∫𝑈𝜀 𝑢

Because 𝜁 is closed, there is a holomorphic function ℎ on 𝑈 such that 𝜁|𝑈 = 𝑑ℎ,

so Stokes’s theorem gives

𝑑𝑢 𝑑𝑢 𝑑𝑢
∧𝜁 = ∧ 𝑑ℎ = − 𝑑 (ℎ )
∫𝑈𝜀 𝑢 ∫𝑈𝜀 𝑢 ∫𝑈𝜀 𝑢
𝑑𝑢 𝑑𝑢 𝑑𝑢
=− ℎ = ℎ + ℎ .
∫𝜕𝑈𝜀 𝑢 ∫𝜕𝐷𝜀 (𝑎) 𝑢 ∫𝜕𝐷𝜀 (𝑏) 𝑢

To compute these last integrals, note that on 𝜕𝐷𝜀 (𝑎) and 𝜕𝐷𝜀 (𝑎), we have 𝑑𝑢/𝑢 =
𝑑𝑧/(𝑧 − 𝑏) − 𝑑𝑧/(𝑧 − 𝑎). We parametrize 𝜕𝐷𝜀 (𝑎) by 𝑧 = 𝑎 + 𝜀𝑒𝑖𝜃 for 𝜃 ∈ [0, 2𝜋],
which yields

𝑑𝑢 𝜀𝑖𝑒𝑖𝜃 𝑑𝜃
2𝜋
ℎ = ℎ(𝑎 + 𝜀𝑒𝑖𝜃 ) − 𝑖 𝑑𝜃 .
∫𝜕𝐷𝜀 (𝑎) 𝑢 ∫0 ( 𝑎 − 𝑏 + 𝜀𝑒𝑖𝜃 )

As 𝜀 → 0, this approaches −2𝜋𝑖ℎ(𝑎). Similarly, the integral over 𝜕𝐷𝜀 (𝑏) ap-
proaches 2𝜋𝑖ℎ(𝑏). Putting these results together, we obtain ∫𝑀 𝛼 ∧ 𝜁 = 2𝜋𝑖(ℎ(𝑏) −
ℎ(𝑎)). But our hypothesis is that 0 = ∫𝛾 𝜁 = ℎ(𝑏) − ℎ(𝑎) for each such 𝜁, so it
follows that 𝛼 ⟂ Ker Δ𝜕 .
The Fredholm theorem then shows that there is some smooth (0, 1)-form 𝛽 such
that 𝛼 = Δ𝜕 𝛽, which is equal to 𝜕𝜕 ∗ 𝛽 because 𝜕 = 0 on (0, 1)-forms. Let 𝑣 = 𝜕 ∗ 𝛽
and 𝑓 = 𝑒−𝑣 𝑢 on 𝑀 ∖ {𝑝, 𝑞}. Then 𝜕𝑓 = 𝑒−𝑣 (𝜕𝑢 − 𝑢𝜕𝑣) = 0, so 𝑓 is holomorphic
away from 𝑝 and 𝑞. Since it agrees with (𝑧 − 𝑎)/(𝑧 − 𝑏) in a neighborhood of 𝑝 and
𝑞, it is meromorphic on 𝑀 and its divisor is equal to 𝑝 − 𝑞.
308 9. Hodge Theory

It remains only to consider the general case in which the image of 𝛾 is not

0 = 𝑡0 < 𝑡1 < ⋯ < 𝑡𝑘 = 1 such that the image of 𝛾𝑗 = 𝛾|[𝑡𝑗−1 ,𝑡𝑗 ] is contained in a
necessarily contained in a coordinate disk. In that case, we can choose numbers

coordinate disk. The construction above yields a function 𝑢𝑗 for each 𝑗 that has a
zero at 𝛾𝑗 (𝑡𝑗−1 ) and a pole at 𝛾𝑗 (𝑡𝑗 ); but 𝛼𝑗 = 𝜕𝑢𝑗 /𝑢𝑗 no longer satisfies ∫𝑀 𝛼𝑗 ∧𝜁 = 0
for holomorphic 1-forms. Instead, the calculation above shows that ∫𝑀 𝛼𝑗 ∧ 𝜁 =
2𝜋𝑖(ℎ(𝛾𝑗 (𝑡𝑗 )) − ℎ(𝛾𝑗 (𝑡𝑗−1 ))) = 2𝜋𝑖 ∫𝛾 𝜁 . Letting 𝑢 denote the product of the 𝑢𝑗 ’s
and 𝛼 = 𝜕𝑢/𝑢, we see that the intermediate poles and zeros cancel, and we have
𝑗

∫𝑀 𝛼 ∧ 𝜁 = ∑𝑗 ∫𝑀 𝛼𝑗 ∧ 𝜁 = ∑𝑗 ∫𝛾 𝜁 = ∫𝛾 𝜁 = 0. The previous argument then

applies to prove the existence of the desired meromorphic function 𝑓 .
𝑗
□

𝑀 of genus 1 can be given the structure of an abelian Lie group, once a particular
It follows from Theorem 9.69 that every connected compact Riemann surface

point 𝑝0 ∈ 𝑀 is chosen to define the map 𝑀 → Jac(𝑀) (so that 𝑝0 becomes the

of genus 1 (sometimes endowed with a specific choice of 𝑝0 , depending on whose

identity in the induced Lie group structure). A connected compact Riemann surface

definition you read) is called an elliptic curve. The name derives from the fact
that elliptic integrals (certain indefinite integrals that appear in the computation of

century to show that nonsingular cubic curves in ℂℙ2 are biholomorphic to complex
the arc length of an ellipse) were used by Abel and Jacobi in the early nineteenth

tori (Problem 10-4). (See [Jos06, Section 5.10] for a discussion of this.) Just for

rather a complex curve in ℂℙ2 with the same equation as an ellipse) is a nonsingular
the record, it should be noted that an ellipse is not an elliptic curve: an ellipse (or

quadric, which as we have seen is always biholomorphic to ℂℙ1 .

Problems
9-1. Suppose 𝑃 is a constant-coefficient second-order scalar differential oper-
ator acting on smooth real-valued functions on ℝ2 . Show that 𝑃 is elliptic
if and only if there is a linear change of coordinates that transforms 𝑃 into
±Δ plus lower-order terms, where Δ is the Laplace–Beltrami operator
with respect to the Euclidean metric.
9-2. Prove Proposition 9.28 (principal symbols of 𝜕, 𝜕 ∗ , and Δ𝜕 ).
9-3. Let (𝑀, 𝑔) be a Kähler manifold. Prove that in every holomorphic coor-
dinate chart, the Laplace–Beltrami operator on scalar functions is given
by the following formula:

𝜕2𝑢
Δ𝑢 = 𝑔 𝑗𝑘 .
𝜕𝑧𝑗 𝜕𝑧𝑘
Problems 309

9-4. Let (𝑀, 𝑔) be a Kähler manifold, and let 𝜔 be its Kähler form, 𝜌 its Ricci
form, and 𝑆 its scalar curvature. Let 𝐿𝜔 be the Lefschetz operator and 𝐿∗𝜔

(a) Show that 𝐿∗𝜔 ∶ ℰ 1,1 (𝑀) → ℰ 0,0 (𝑀) has the coordinate formula
its adjoint.

𝐿∗𝜔 (𝑖𝜂𝑗𝑘 𝑑𝑧𝑗 ∧ 𝑑𝑧𝑘 ) = 𝑔 𝑗𝑘 𝜂𝑗𝑘 .

(b) Use part (a) to derive the following formulas:

𝐿∗𝜔 𝜔 = dim 𝑀,
𝐿∗𝜔 𝜌 = 21 𝑆,
𝐿∗𝜔 (𝑖𝜕𝜕𝑢) = Δ𝑢 for 𝑢 ∈ 𝐶 ∞ (𝑀; ℂ).

9-5. Let (𝑀, 𝑔) be a Kähler manifold and 𝜔 its Kähler form. Show that 𝜔 is
harmonic.
9-6. A cohomology class 𝛾 ∈ 𝐻dR 𝑘
(𝑀; ℂ)) on an 𝑛-dimensional Kähler mani-
fold 𝑀 is said to be primitive if 𝐿𝑛−𝑘+1
𝜔 𝛾 = 0, where 𝐿𝜔 is the Lefschetz
operator. Let 𝑃 𝑘 (𝑀) ⊆ 𝐻dR𝑘
(𝑀; ℂ) denote the space of primitive degree-
𝑘 cohomology classes, and 𝑃 𝑝,𝑞 (𝑀) ⊆ 𝐻 𝑝,𝑞 (𝑀) the space of primitive
(𝑝, 𝑞)-classes. Prove the Lefschetz decomposition theorem: Let 𝑀 be a
compact 𝑛-dimensional Kähler manifold. There are direct sum decompo-
sitions

𝐻dR
𝑘
(𝑀; ℂ) = 𝐿𝑟𝜔 𝑃 𝑘−2𝑟 (𝑀),
⨁
0≤𝑟≤𝑘/2

𝐻 𝑝,𝑞
(𝑀) = 𝐿𝑟𝜔 𝑃 𝑝−𝑟,𝑞−𝑟 (𝑀).
⨁
0≤𝑟≤(𝑝+𝑞)/2

[Hint: Use the hard Lefschetz theorem and induction on 𝑘.]

9-7. Let (𝑀, 𝑔) be a Riemannian manifold.
(a) Prove that every harmonic 1-form 𝜂 on 𝑀 satisfies the following
identity:

Δ|𝜂|2 = |∇𝜂|2 + ⟨𝑅𝑐 ♯ (𝜂), 𝜂⟩,

where 𝑅𝑐 ♯ ∶ Λ1 𝑀 → Λ1 𝑀 is the bundle homomorphism given by

𝑅𝑐 ♯ (𝜂)(𝑣) = 𝑅𝑐(𝜂 ♯ , 𝑣). [Hint: Do the computation in Riemannian
normal coordinates.]

due to Salomon Bochner [Boc46]: If (𝑀, 𝑔) is a compact Riemann-

(b) Use the result of part (a) to prove the Bochner vanishing theorem,

ian 𝑛-manifold with nonnegative Ricci curvature, then 𝑏1 (𝑀) ≤

𝑛. If in addition the Ricci curvature is positive somewhere, then
𝑏1 (𝑀) = 0.
310 9. Hodge Theory

9-8. Let (𝑀, 𝑔) be a Kähler manifold.

(a) Prove that every holomorphic 𝑝-form 𝜂 on 𝑀 satisfies the following
identity:
Δ|𝜂|2 = |∇𝜂|2 + ⟨𝑅𝑐 ♯ (𝜂), 𝜂⟩,
where 𝑅𝑐 ♯ ∶ Λ𝑝,0 𝑀 → Λ𝑝,0 𝑀 is the bundle endomorphism given in
holomorphic coordinates by
𝑝
𝑅𝑐 (𝜂)𝑗1 …𝑗𝑝 =
♯
𝑔 𝑚𝑘 𝑅𝑗𝑠 𝑘 𝜂𝑗1 …𝑚…𝑗𝑝 .
∑
𝑠=1

(8.8) at an arbitrary point 𝑎, and use the result of Problem 8-11.]

[Hint: One way to do this is to compute in coordinates satisfying

(b) Prove the following theorem, also due to Bochner [Boc46]: If (𝑀, 𝑔)

every holomorphic differential form on 𝑀 is parallel. If in addition

is a compact Kähler manifold with nonnegative Ricci curvature, then

the Ricci curvature is positive somewhere, then ℎ𝑝,0 (𝑀) = 0 for all
𝑝 > 0.
9-9. Prove that every automorphism of ℂℙ𝑛 is a projective transformation.
[Hint: Given an automorphism 𝐹 ∶ ℂℙ𝑛 → ℂℙ𝑛 , prove that 𝐹 ∗ 𝑇 ≅ 𝑇 ,
where 𝑇 → ℂℙ𝑛 is the tautological bundle, and use this fact to construct
a holomorphic map 𝐹̃ ∶ 𝑇 → 𝑇 , linear on fibers, such that the following
diagram commutes:
𝐹̃ /
𝑇 𝑇

ℂℙ𝑛 / ℂℙ𝑛 .]
𝐹
 

9-10. Prove that a holomorphic line bundle on a compact complex manifold

cannot be both positive and negative.
9-11. Show that there is an orientation-reversing diffeomorphism 𝐹 ∶ ℂℙ𝑛 →
ℂℙ𝑛 if and only if 𝑛 is odd. Together with the result of Problem 3-10,
this shows that the blowup of a complex 𝑛-manifold 𝑀 at a point is dif-
feomorphic to 𝑀 # ℂℙ𝑛 when 𝑛 is odd, but not necessarily when 𝑛 is

𝐹 ∗ (𝜔𝑛 ), where 𝜔 is a Kähler form on ℂℙ𝑛 .]

even. [Hint: In the even case, consider the de Rham cohomology class of

9-12. Suppose (𝑀, 𝑔) is a compact Kähler manifold with constant scalar cur-
vature whose Kähler class is equal to a multiple of 𝑐1ℝ (𝑀). Prove that
𝑀 is Kähler-Einstein. [Hint: Show that a constant multiple of the Ricci
form is equal to 𝜔 + 𝑖𝜕𝜕𝑢 for some scalar function 𝑢, and use the result of
Problem 9-4 to conclude that Δ𝑢 is constant. Conclude from this that 𝑢 is
constant.]
Problems 311

9-13. Suppose 𝑇 = 𝑉 /Λ is a complex torus of dimension 𝑛. Define the dual

torus to be the group 𝑇 ∗ = Hom(𝑇 , U(1)) (the set of homomorphisms
from 𝑇 to the circle group, which is a group under pointwise multiplica-

(a) Let 𝑉 ∗ denote the complex vector space of conjugate-linear func-

tion).

tionals from 𝑉 to ℂ. Show there is a surjective group homomor-

phism Φ ∶ 𝑉 ∗ → 𝑇 ∗ that sends 𝑓 to the homomorphism Φ(𝑓 )(𝑣) =
𝑒2𝜋𝑖 Im 𝑓 (𝑣) , and its kernel is the lattice
Λ∗ = {𝑓 ∈ 𝑉 ∗ ∶ Im 𝑓 (𝑣) ∈ ℤ for all 𝑣 ∈ Λ}.
Conclude that 𝑇 ∗ also has the structure of a complex torus of dimen-
sion 𝑛, isomorphic to 𝑉 ∗ /Λ∗ .
(b) For a compact Kähler manifold 𝑀, show that Alb(𝑀) is holomor-
phically isomorphic to the dual torus of Pic0 (𝑀).
9-14. Let 𝑀 be a connected compact Riemann surface of genus 𝑔 ≥ 1, and
let Div0 (𝑀) be the group of divisors of degree 0 on 𝑀. Define a map
𝜇 ∶ Div0 (𝑀) → Jac(𝑀) = Ω1 (𝑀)∗ /Λ by choosing a point 𝑝0 ∈ 𝑀 and
setting
𝑞𝑗
𝜇 ( 𝑛𝑗 𝑞𝑗 ) = (𝜁 ↦ 𝑛 𝜁) mod Λ.
∑ ∑ 𝑗∫𝑝
𝑗 𝑗 0

Prove Abel’s theorem: A divisor 𝐷 ∈ Div0 (𝑀) is a principal divisor if

and only if 𝜇(𝐷) = 0. [Hint: For the “only if” part, assume 𝐷 = (𝑓 ) and
construct a map Φ ∶ ℂℙ1 → Jac(𝑀) by Φ([𝑡0 , 𝑡1 ]) = 𝜇 ((𝑡0 𝑓 + 𝑡1 )); then
use the result of Problem 4-6 to show Φ is constant. For the “if” part, adapt
the proof of Lemma 9.70.] [Remark: This was Abel’s half of the original

the map 𝜇 defined above descends to an injective homomorphism from

proof of the Abel–Jacobi theorem (Thm. 9.68). Abel’s theorem shows that

Cl0 (𝑀) (which is isomorphic to Pic0 (𝑀) by Corollary 9.63) to Jac(𝑀).

The other half of the theorem, surjectivity of 𝜇, is known as the Jacobi
inversion theorem. For a proof, see [GH94, p. 235].]
9-15. Let 𝑀 be a connected compact Riemann surface of genus 𝑔 ≥ 1. Prove
that the canonical bundle 𝐾 → 𝑀 has no base points. [Hint: Assuming
𝑝 is a base point, show that 𝒪(𝐾) ≅ ℐ{𝑝} (𝐾) ≅ 𝒪(𝐾 ⊗ 𝐿∗{𝑝} ), and use
Riemann–Roch to calculate dim 𝒪(𝑀; 𝐿{𝑝} ).]
9-16. Let 𝑀 be a compact Kähler manifold. Define a map 𝐴𝑀 ∶ 𝑀 → Alb(𝑀),
called the Albanese map, by choosing a point 𝑝0 ∈ 𝑀 and setting
𝑞
𝐴𝑀 (𝑞) = (𝜁 ↦ 𝜁) mod Λ.
∫𝑝0

(This generalizes the map defined in the proof of Theorem 9.69.)

312 9. Hodge Theory

(a) Show that 𝐴𝑀 is holomorphic.

(b) Show that if 𝑀 is a compact Riemann surface of genus 𝑔 ≥ 1, then
𝐴𝑀 is a holomorphic embedding.
(c) Show that if 𝑀 = ℂ𝑛 /Λ0 is a complex torus, then 𝐴𝑀 is a biholo-

(d) Let 𝐹 ∶ 𝑀 → 𝑁 be a holomorphic map between compact Kähler

morphism.

manifolds. Show that there is a holomorphic map 𝐹̃ ∶ Alb(𝑀) →

Alb(𝑁) such that the following diagram commutes:

𝐹
𝑀 /𝑁

𝐴𝑀 𝐴𝑁
 ̃
𝐹 / 
Alb(𝑀) Alb(𝑁).

[Hint: Begin by defining 𝐹̂ ∶ Ω1 (𝑀)∗ → Ω1 (𝑁)∗ by 𝐹̂(𝜑)(𝜂) =

𝜑(𝐹 ∗ 𝜂). You will need to choose the base points for the Albanese

(e) Prove the universal property of the Albanese variety: If 𝑀 is a

maps carefully.]

compact Kähler manifold and 𝐹 ∶ 𝑀 → 𝑇 is a holomorphic map

to a complex torus, then 𝐹 factors through the Albanese variety of
𝑀; that is, there is a holomorphic map 𝜓 ∶ Alb(𝑀) → 𝑇 such that
𝐹 = 𝜓 ∘ 𝐴𝑀 .
9-17. Suppose 𝐸, 𝐹 are smooth Hermitian vector bundles on a Riemannian
manifold 𝑀, and 𝑃 ∶ Γ(𝐸) → Γ(𝐹 ) is a differential operator. We say
𝑃 has injective symbol if the principal symbol map 𝜎𝑃 (𝑥, 𝜉) ∶ 𝐸𝑥 → 𝐹𝑥
is injective for every 𝑥 ∈ 𝑀 and every nonzero 𝜉 ∈ 𝑇𝑥∗ 𝑀. Prove the
following Fredholm theorem for operators with injective symbol: If 𝑀
is compact and 𝑃 ∶ Γ(𝐸) → Γ(𝐹 ) has injective symbol, then Ker 𝑃 is
finite-dimensional and there is an orthogonal direct sum decomposition
Γ(𝐸) = Ker 𝑃 ⊕ Im 𝑃 ∗ .
[Hint: Consider the principal symbol of 𝑃 ∗ 𝑃 .]
9-18. Let 𝑀 be a Riemannian manifold and 𝐸0 , 𝐸1 , … a sequence of Hermitian
vector bundles on 𝑀. A sequence of first-order linear differential opera-
tors
𝑃0 𝑃1 𝑃2
(9.55) 0 → Γ(𝐸0 ) ⟶ Γ(𝐸1 ) ⟶ Γ(𝐸2 ) ⟶ ⋯
is called an elliptic complex if for each 𝑗, 𝑃𝑗+1 ∘ 𝑃𝑗 = 0 and the principal
symbol sequence
𝜎𝑃𝑗 (𝑥,𝜉) 𝜎𝑃𝑗+1 (𝑥,𝜉)
⋯ → 𝐸𝑗 |𝑥 −−−−−−→ 𝐸𝑗+1 |𝑥 −−−−−−−→ 𝐸𝑗+2 |𝑥 → ⋯
Problems 313

is exact for every 𝑥 ∈ 𝑀 and every nonzero 𝜉 ∈ 𝑇𝑥∗ 𝑀. Given such a

complex, define
ℋ 𝑗 (𝐸∗ ) = {𝜎 ∈ Γ(𝐸𝑗 ) ∶ 𝑃𝑗 𝜎 = 0 and 𝑃𝑗−1
∗
𝜎 = 0};
Ker (𝑃𝑗 ∶ Γ(𝐸𝑗 ) → Γ(𝐸𝑗+1 ))
𝐻 𝑗 (𝐸∗ ) = .
Im (𝑃𝑗−1 ∶ Γ(𝐸𝑗−1 ) → Γ(𝐸𝑗 ))
Prove the following Hodge theorem for elliptic complexes: If 𝑀 is com-
pact and (9.55) is an elliptic complex on 𝑀, then for every 𝑗, ℋ 𝑗 (𝐸∗ ) is
finite-dimensional, and we have an orthogonal direct sum decomposition
Γ(𝐸𝑗 ) = ℋ 𝑗 (𝐸∗ ) ⊕ Im 𝑃𝑗−1 ⊕ Im 𝑃𝑗+1
∗
,
and an isomorphism
𝐻 𝑗 (𝐸∗ ) ≅ ℋ 𝑗 (𝐸∗ ).
[Hint: First show that 𝑃𝑗 ⊕ 𝑃𝑗−1
∗
∶ Γ(𝐸𝑗 ) → Γ(𝐸𝑗+1 ⊕ 𝐸𝑗−1 ) has injective
symbol.]
 10.1090/gsm/244/10

Chapter 10

The Kodaira Embedding

Theorem

In this chapter, we present the proof of the Kodaira embedding theorem, which
completely characterizes those compact complex manifolds that can be holomor-
phically embedded in projective spaces. The theorem comes in two versions: The
first, more technical version states that a compact complex manifold is projective

class is represented by a positive (1, 1)-form. The second, more geometric version
if and only if it carries a positive line bundle, meaning one whose first real Chern

states that a compact complex manifold is projective if and only if it admits a Käh-
ler metric whose Kähler class is integral, meaning it lies in the image of integral
cohomology under the coefficient homomorphism.
The proof of this theorem uses most of the tools we have developed in this book:
blowups, Hartogs’s theorem, holomorphic line bundles, Chern connections, sheaf
cohomology, and Hodge theory. It was first published in 1954 by Kunihiko Kodaira
[Kod54].

Preliminaries
The heart of Kodaira’s embedding theorem is the statement that every positive holo-
morphic line bundle over a compact complex manifold is ample, meaning that some
positive tensor power of it is very ample: its global holomorphic sections separate
points and directions. It then follows from Corollary 3.44 that a compact complex

Let us consider an arbitrary holomorphic line bundle 𝐿 → 𝑀, and think about

manifold carrying a positive line bundle is projective.

how one might go about proving it is very ample. We need to show that 𝒪(𝑀; 𝐿)
separates points and separates directions.

315
316 10. The Kodaira Embedding Theorem

sections of sheaves. Thus let 𝑝 and 𝑞 be a pair of distinct points in 𝑀, and let
The first step is to transform the desired conclusions into statements about

(𝐿𝑝 )𝑝 ⊕ (𝐿𝑞 )𝑞 be the “double skyscraper sheaf” whose stalks at 𝑝 and 𝑞 are the
fibers 𝐿𝑝 and 𝐿𝑞 , respectively, and all other stalks are zero. There is a sheaf mor-
phism 𝑒 ∶ 𝒪(𝐿) → (𝐿𝑝 )𝑝 ⊕(𝐿𝑞 )𝑞 called the evaluation map, defined as follows: for
𝜎 ∈ 𝒪(𝑈 ; 𝐿) over some open set 𝑈 ⊆ 𝑀, we set 𝑒𝑈 (𝜎) = (𝜎(𝑝), 𝜎(𝑞)) ∈ 𝐿𝑝 ⊕ 𝐿𝑞
if 𝑝 and 𝑞 both lie in 𝑈 ; if only one of them does, then 𝑒𝑈 (𝜎) = 𝜎(𝑝) or 𝜎(𝑞) as
appropriate; and otherwise 𝑒𝑈 (𝜎) = 0. Then the statement that 𝒪(𝑀; 𝐿) separates
points is equivalent to the global section map 𝑒𝑀 ∶ 𝒪(𝑀; 𝐿) → 𝐿𝑝 ⊕ 𝐿𝑞 being
surjective for all 𝑝 and 𝑞.
The evaluation map fits into a short exact sheaf sequence:

𝑒
(10.1) 0 → ℐ{𝑝,𝑞} (𝐿) ↪ 𝒪(𝐿) → (𝐿𝑝 )𝑝 ⊕ (𝐿𝑞 )𝑞 → 0,

where ℐ{𝑝,𝑞} (𝐿) is the sheaf of holomorphic sections of 𝐿 that vanish at 𝑝 and 𝑞.

section map 𝑒𝑀 being surjective lies in the cohomology group 𝐻 1 (𝑀; ℐ{𝑝,𝑞} (𝐿)).
It is easy to check that this sheaf sequence is exact, so the obstruction to the global

Similarly, the question of whether 𝒪(𝑀; 𝐿) separates directions can also be

reframed as a question about sheaves. Let 𝑝 be an arbitrary point in 𝑀, and choose
a local holomorphic frame 𝑠 for 𝐿 in a neighborhood of 𝑝. Let ℐ{𝑝} (𝐿) be the sheaf
of holomorphic sections of 𝐿 that vanish at 𝑝, and let ℐ{𝑝}2
(𝐿) be the subsheaf of
ℐ{𝑝} (𝐿) consisting of sections that vanish to second order at 𝑝 (see Example 5.2 for
the definition). Then we have an exact sheaf sequence

𝛿
0 → ℐ{𝑝} (𝐿) ↪ ℐ{𝑝} (𝐿) → (Λ𝑝 𝑀)𝑝 → 0,
2 1,0
(10.2)

where the sheaf on the right is the skyscraper sheaf whose stalk at 𝑝 is Λ𝑝 𝑀 with
1,0

all other stalks zero, and 𝛿(𝑓 𝑠) = 𝑑𝑓𝑝 . Surjectivity at (Λ𝑝 𝑀)𝑝 can be proved
1,0

by choosing holomorphic coordinates (𝑧1 , … , 𝑧𝑛 ) on a neighborhood 𝑈 of 𝑝 and

centered at 𝑝, and noting that for any 𝜂 = ∑𝑗 𝑐𝑗 𝑑𝑧𝑗 |𝑝 ∈ Λ𝑝 𝑀, the section 𝜎(𝑧) =
1,0

( ∑𝑗 𝑐𝑗 𝑧 )𝑠(𝑧) lies in ℐ{𝑝} (𝑈 ; 𝐿) and satisfies 𝛿(𝜎) = 𝜂. Exactness at ℐ{𝑝} (𝐿)

𝑗

follows from the fact that, thanks to Taylor’s theorem, a holomorphic function 𝑓
satisfying 𝑓 (𝑝) = 0 and 𝑑𝑓𝑝 = 0 can be written in local holomorphic coordinates
centered at 𝑝 in the form 𝑓 (𝑧) = ∑𝑗,𝑘 𝑧𝑗 𝑧𝑘 𝑔𝑗𝑘 (𝑧) for some holomorphic functions
𝑔𝑗𝑘 .
The next exercise shows that proving 𝒪(𝑀; 𝐿) separates directions is equiva-
lent to proving that the global section map 𝛿𝑀 ∶ ℐ{𝑝} (𝑀; 𝐿) → Λ𝑝 𝑀 is surjec-
1,0

tive for every 𝑝 ∈ 𝑀. Once again, the obstruction lies in the cohomology group
𝐻 1 (𝑀; ℐ{𝑝}
2
(𝐿)).
The Kodaira Vanishing Theorem 317

► Exercise 10.1. With notation as above, show that 𝛿𝑀 ∶ ℐ{𝑝} (𝑀; 𝐿) → Λ1,0
𝑝 𝑀
is surjective if and only if for every nonzero 𝑣 ∈ 𝑇𝑝′ 𝑀, there exists a global section
𝜎 ∈ ℐ{𝑝} (𝑀; 𝐿) that can be expressed in a neighborhood of 𝑝 as 𝜎 = 𝑓 𝑠, where
𝑓 (𝑝) = 0 and 𝑣𝑓 ≠ 0.

homology of sheaves like ℐ{𝑝,𝑞} (𝐿) or ℐ{𝑝} 2

(𝐿) (except in the case dim 𝑀 = 1, as
This setup is promising, but unfortunately we have no tools for evaluating co-

we will see below). On the other hand, the sheaf of sections of 𝐿 that vanish on
a hypersurface is isomorphic to the sheaf of sections of a holomorphic line bundle
by the result of Proposition 5.16, and we will prove a theorem below (the Kodaira

such a sheaf are zero. So the plan is to transfer the problem to the blowup of 𝑀 at
vanishing theorem) that gives conditions under which the cohomology groups of

the points 𝑝 and 𝑞 (for separating points) or at 𝑝 alone (for separating directions),
where the selected points are replaced by hypersurfaces, and try to find an appro-

dim 𝑀 = 1, individual points are hypersurfaces, so blowing up is not necessary.

priate line bundle to which we can apply the Kodaira vanishing theorem. (When

As Problem 10-3 shows, the proof is simpler in that case.)

The Kodaira Vanishing Theorem

The main tool for proving that the crucial cohomology groups are zero is a vanishing
theorem for certain cohomology groups of sheaves of sections of line bundles. We
begin by extending the Hodge-theoretic results of Chapter 9 to bundle-valued forms

Suppose 𝑀 is a complex manifold and 𝐸 → 𝑀 is a Hermitian vector bundle.

on a Kähler manifold.

Let ∇ be a connection on 𝐸, and let 𝐷 ∶ ℰ 𝑘 (𝑀; 𝐸) → ℰ 𝑘+1 (𝑀; 𝐸) be the exterior

covariant derivative determined by ∇ (Prop. 7.11). On (𝑝, 𝑞)-forms, we can decom-
pose this operator as 𝐷 = 𝐷′ + 𝐷″ , where 𝐷′ maps ℰ 𝑝,𝑞 (𝑀; 𝐸) to ℰ 𝑝+1,𝑞 (𝑀; 𝐸)
and 𝐷″ maps ℰ 𝑝,𝑞 (𝑀; 𝐸) to ℰ 𝑝,𝑞+1 (𝑀; 𝐸).
We define two Laplace-type operators Δ′ , Δ″ ∶ ℰ 𝑝,𝑞 (𝑀; 𝐸) → ℰ 𝑝,𝑞 (𝑀; 𝐸):
(10.3) Δ′ 𝛼 = 𝐷′ 𝐷′∗ 𝛼 + 𝐷′∗ 𝐷′ 𝛼, Δ″ 𝛼 = 𝐷″ 𝐷″∗ 𝛼 + 𝐷″∗ 𝐷″ 𝛼.
Unlike the case of the Dolbeault Laplacian and its conjugate acting on scalar-valued
forms, these two operators are generally not equal on Kähler manifolds; but in the
case of a holomorphic bundle with its Chern connection, there is an important for-
mula for their difference. The first step in relating these operators is to prove a
bundle-valued analogue of the Kähler identities.
Proposition 10.2 (Kähler Identities for Bundle-Valued Forms). Suppose 𝑀 is
a Kähler manifold with Kähler form 𝜔; 𝐸 → 𝑀 is a smooth Hermitian vector
bundle; ∇ is a metric connection on 𝐸; 𝐷′ , 𝐷″ are the exterior covariant derivative
operators defined with respect to ∇;and 𝐿𝜔 ∶ ℰ 𝑝,𝑞 (𝑀; 𝐸) → ℰ 𝑝+1,𝑞+1 (𝑀; 𝐸) is the
318 10. The Kodaira Embedding Theorem

bundle homomorphism 𝐿𝜔 𝜂 = 𝜔 ∧ 𝜂. The following identities hold:

(a) [𝐷″∗ , 𝐿𝜔 ] = 𝑖𝐷′ .
(b) [𝐷′∗ , 𝐿𝜔 ] = −𝑖𝐷″ .
(c) [𝐿∗𝜔 , 𝐷″ ] = −𝑖𝐷′∗ .
(d) [𝐿∗𝜔 , 𝐷′ ] = 𝑖𝐷″∗ .

Proof. We begin with (a). Suppose (𝑠𝑗 ) is a local frame for 𝐸 on an open set
𝑈 ⊆ 𝑀, and 𝛼 = 𝛼 𝑗 ⊗ 𝑠𝑗 is an 𝐸-valued (𝑝, 𝑞 − 1)-form. If 𝜃𝑗𝑘 denotes the matrix
of connection 1-forms with respect to this frame, then (7.11) gives the following
local expression for 𝐷𝛼:

𝐷𝛼 = 𝑑𝛼 𝑗 ⊗ 𝑠𝑗 + (−1)𝑝+𝑞−1 (𝛼 𝑗 ∧ 𝜃𝑗𝑘 ) ⊗ 𝑠𝑘 ,

and therefore

𝐷′ 𝛼 = 𝜕𝛼 𝑗 ⊗ 𝑠𝑗 + (−1)𝑝+𝑞−1 (𝛼 𝑗 ∧ (𝜃 ′ )𝑘𝑗 ) ⊗ 𝑠𝑘 ,
𝐷″ 𝛼 = 𝜕𝛼 𝑗 ⊗ 𝑠𝑗 + (−1)𝑝+𝑞−1 (𝛼 𝑗 ∧ (𝜃 ″ )𝑘𝑗 ) ⊗ 𝑠𝑘 ,

where 𝜃 ′ and 𝜃 ″ represent the (1, 0)- and (0, 1)-parts of 𝜃, respectively. Let us write
the second terms on the right-hand sides as 𝜃 ′ 𝛼 and 𝜃 ″ 𝛼, so

𝐷′ 𝛼 = 𝜕𝛼 𝑗 ⊗ 𝑠𝑗 + 𝜃 ′ 𝛼,
𝐷″ 𝛼 = 𝜕𝛼 𝑗 ⊗ 𝑠𝑗 + 𝜃 ″ 𝛼,

Now suppose 𝛼 is as above and 𝛽 = 𝛽 𝑘 ⊗ 𝑠𝑘 is an 𝐸-valued (𝑝, 𝑞)-form, and

both are compactly supported in 𝑈 . Then by the way we have defined the inner
product of 𝐸-valued forms and the fact that the frame (𝑠𝑗 ) is orthonormal, we have

(𝐷″ 𝛼, 𝛽) = ⟨𝜕𝛼 𝑗 ⊗ 𝑠𝑗 , 𝛽 𝑘 ⊗ 𝑠𝑘 ⟩𝑑𝑉𝑔 + (𝜃 ″ 𝛼, 𝛽)

∫𝑀

= ⟨𝜕𝛼 𝑗 , 𝛽 𝑗 ⟩𝑑𝑉𝑔 + (𝜃 ″ 𝛼, 𝛽)
∫ ∑
𝑀 𝑗

= ⟨𝛼 𝑗 , 𝜕 ∗ 𝛽 𝑗 ⟩𝑑𝑉𝑔 + (𝛼, 𝜃 ″∗ 𝛽),

∫ ∑
𝑀 𝑗

which shows that

𝐷″∗ 𝛽 = 𝜕 ∗ 𝛽 𝑘 ⊗ 𝑠𝑘 + 𝜃 ″∗ 𝛽

form compactly supported in 𝑈 , this expression holds everywhere in 𝑈 .

for such forms. Since every form agrees on a neighborhood of each point with a
The Kodaira Vanishing Theorem 319

Since 𝐿𝜔 𝛽 = (𝐿𝜔 𝛽 𝑘 ) ⊗ 𝑠𝑘 , we have

𝐷″∗ 𝐿𝜔 𝛽 = 𝐷″∗ ((𝐿𝜔 𝛽 𝑘 ) ⊗ 𝑠𝑘 )
= (𝜕 ∗ 𝐿𝜔 𝛽 𝑘 ) ⊗ 𝑠𝑘 + 𝜃 ″∗ 𝐿𝜔 𝛽,
and
𝐿𝜔 𝐷″∗ 𝛽 = 𝐿𝜔 (𝜕 ∗ 𝛽 𝑘 ) ⊗ 𝑠𝑘 + 𝐿𝜔 𝜃 ″∗ 𝛽.
Therefore, using the corresponding Kähler identity for scalar-valued forms (Prop.
9.38(a)), we find
[𝐷″∗ , 𝐿𝜔 ]𝛽 = ([𝜕 ∗ , 𝐿𝜔 ]𝛽 𝑘 ) ⊗ 𝑠𝑘 + [𝜃 ″∗ , 𝐿𝜔 ]𝛽
(10.4) = (𝑖𝜕𝛽 𝑘 ) ⊗ 𝑠𝑘 + [𝜃 ″∗ , 𝐿𝜔 ]𝛽
= 𝑖𝐷′ 𝛽 − 𝑖𝜃 ′ 𝛽 + [𝜃 ″∗ , 𝐿𝜔 ]𝛽.

This formula holds with respect to every local orthonormal frame. On the other

each 𝑥0 ∈ 𝑀 we can use any convenient frame to verify the identity at 𝑥0 . Lemma
hand, since both sides of (a) are defined independently of any choice of frame, for

7.3 shows that we can choose an orthonormal frame with the property that 𝜃𝑘 = 0
𝑗

at 𝑥0 . Because 𝜃 and 𝜃 do not involve differentiation, it follows that 𝜃 and 𝜃 ″∗

′ ″ ′

both vanish at 𝑥0 , and then (10.4) shows that (a) holds at 𝑥0 . This proves (a).
The proof of (b) is exactly the same, with single-primed and double-primed
expressions interchanged; and then (c) and (d) follow by taking adjoints. □

Besides the Kähler identities, the other key ingredient in proving that Δ𝜕 = Δ𝜕
on scalar-valued forms (Lemma 9.42) was the fact that 𝜕𝜕 = −𝜕𝜕, which follows
immediately from 𝑑 2 = 0. But in the present context, the role of 𝑑 is played by
the exterior covariant derivative 𝐷, whose square is not zero. Instead, we have the
following lemma.
Lemma 10.3. Suppose 𝐸 → 𝑀 is a Hermitian holomorphic vector bundle and
∇ is its Chern connection. Let 𝐷 = 𝐷′ + 𝐷″ be the exterior covariant derivative
associated with ∇, and let Θ ∈ ℰ 1,1 (𝑀; 𝐸) be its curvature. Then for every 𝐸-
valued differential form 𝛼,
𝐷′ 𝐷″ 𝛼 + 𝐷″ 𝐷′ 𝛼 = Θ ∧ 𝛼.

Proof. For 𝛼 ∈ ℰ 𝑝,𝑞 (𝑀; 𝐸), Proposition 7.11(iii) shows that

Θ ∧ 𝛼 = 𝐷2 𝛼 = (𝐷′ )2 𝛼 + (𝐷′ 𝐷″ 𝛼 + 𝐷″ 𝐷′ 𝛼 ) + (𝐷″ )2 𝛼.
Because Θ is of type (1, 1) (Prop. 7.18), the first and last terms on the right-hand
side are zero, and the result follows. □

ators Δ′ and Δ″ , due to Yasuo Akizuki and Shigeo Nakano [AN54].

Here is the fundamental relation between the two bundle-valued Laplace oper-
320 10. The Kodaira Embedding Theorem

Theorem 10.4 (Akizuki–Nakano Identity). Suppose 𝑀 is a Kähler manifold and

𝐸 → 𝑀 is a Hermitian holomorphic vector bundle. Let Δ′ and Δ″ be the Laplace
operators defined by (10.3) with respect to the Chern connection on 𝐸, and let
Ω ∶ ℰ 𝑝,𝑞 (𝑀; 𝐸) → ℰ 𝑝+1,𝑞+1 (𝑀; 𝐸) be the operator Ω(𝛼) = 𝑖Θ ∧ 𝛼, where Θ is the
curvature of the Chern connection. Then
Δ″ = Δ′ + [Ω, 𝐿∗𝜔 ].

Proof. Using Proposition 10.2, we compute

Δ″ = 𝐷″ 𝐷″∗ + 𝐷″∗ 𝐷″
= −𝑖𝐷″ [𝐿∗𝜔 , 𝐷′ ] − 𝑖[𝐿∗𝜔 , 𝐷′ ]𝐷″
= −𝑖(𝐷″ 𝐿∗𝜔 𝐷′ − 𝐷″ 𝐷′ 𝐿∗𝜔 + 𝐿∗𝜔 𝐷′ 𝐷″ − 𝐷′ 𝐿∗𝜔 𝐷″ ),
and
Δ′ = 𝐷′ 𝐷′∗ + 𝐷′∗ 𝐷′
= 𝑖(𝐷′ [𝐿∗𝜔 , 𝐷″ ] + [𝐿∗𝜔 , 𝐷″ ]𝐷′ )
= 𝑖(𝐷′ 𝐿∗𝜔 𝐷″ − 𝐷′ 𝐷″ 𝐿∗𝜔 + 𝐿∗𝜔 𝐷″ 𝐷′ − 𝐷″ 𝐿∗𝜔 𝐷′ ).
Therefore, Lemma 10.3 gives
Δ″ − Δ′ = 𝑖(𝐷″ 𝐷′ + 𝐷′ 𝐷″ )𝐿∗𝜔 − 𝑖𝐿∗𝜔 (𝐷″ 𝐷′ + 𝐷′ 𝐷″ )
= [Ω, 𝐿∗𝜔 ]. □

This does not directly yield results analogous to the Hodge decomposition for
scalar-valued forms; but in special cases when something can be said about the
curvature, it can lead to very powerful results. We will see one such application
below—the Kodaira–Nakano–Akizuki vanishing theorem (Thm. 10.6).

The Kodaira–Nakano–Akizuki Vanishing Theorem

Recall that a holomorphic line bundle 𝐿 is said to be positive if its first real
Chern class is represented by a positive (1, 1)-form, that is, a real (1, 1)-form 𝜔
such that 𝜔(𝑋, 𝐽 𝑋) > 0 for all nonzero 𝑋. For example, the hyperplane bundle on
ℂℙ𝑛 is positive because it has a fiber metric whose Chern form is a positive multiple
of the Kähler form of the Fubini–Study metric (see Example 8.14).
Proposition 10.5 (Basic Properties of Positive Line Bundles). Suppose 𝑀 is a
compact complex manifold and 𝐿 → 𝑀 is a positive holomorphic line bundle.
(a) There exist a Kähler metric on 𝑀 and a Hermitian fiber metric on 𝐿 such
that 2𝜋𝑖 Θ𝐿 is equal to the Kähler form, where Θ𝐿 is the curvature of the
Chern connection on 𝐿.
(b) If 𝑀 ′ is another complex manifold and 𝑓 ∶ 𝑀 ′ → 𝑀 is a holomorphic
immersion, then 𝑓 ∗ 𝐿 → 𝑀 ′ is also positive.
The Kodaira Vanishing Theorem 321

Proof. To prove (a), let 𝜔 be a closed, positive (1, 1)-form representing 𝑐1ℝ (𝐿).
Then 𝜔 is a Kähler form, so we can endow 𝑀 with the Kähler metric 𝑔 = 𝜔(⋅, 𝐽 ⋅).
By Corollary 9.58, there is a Hermitian fiber metric on 𝐿 whose Chern connection
has a curvature form Θ𝐿 that satisfies 2𝜋𝑖 Θ𝐿 = 𝜔.
For (b), suppose 𝑓 ∶ 𝑀 ′ → 𝑀 is a holomorphic immersion. Give 𝐿 the Her-
mitian fiber metric described in part (a), and give 𝑓 ∗ 𝐿 the pullback metric. Then
Proposition 7.21 shows that the curvature of the Chern connection on 𝑓 ∗ 𝐿 is given
by Θ𝑓 ∗ 𝐿 = 𝑓 ∗ Θ𝐿 . Because 𝐹 is a holomorphic immersion, it follows that for all
𝑥 ∈ 𝑀 ′ and all nonzero 𝑋 ∈ 𝑇𝑥 𝑀 ′ ,
𝑖 𝑖 ∗
Θ ∗ (𝑋, 𝐽 𝑋) = 𝑓 Θ𝐿 |𝑥 (𝑋, 𝐽 𝑋)
2𝜋 𝑓 𝐿 |𝑥 2𝜋
𝑖
= Θ (𝐷𝑓 (𝑥)(𝑋), 𝐽 𝐷𝑓 (𝑥)(𝑋)) > 0,
2𝜋 𝐿 |𝑓 (𝑥)
which shows that 𝑓 ∗ 𝐿 is positive. □

The following theorem expresses one of the deepest properties of positive line
bundles. It was proved in 1954 by Yasuo Akizuki and Shigeo Nakano [AN54],
generalizing an earlier result by Kunihiko Kodaira (Cor. 10.7 below).
Theorem 10.6 (Kodaira–Nakano–Akizuki Vanishing Theorem). Suppose 𝑀 is
a compact 𝑛-dimensional complex manifold and 𝐿 → 𝑀 is a positive line bundle.
Then 𝐻 𝑝,𝑞 (𝑀; 𝐿) = 0 for 𝑝 + 𝑞 > 𝑛.

Proof. Give 𝑀 the Kähler metric and 𝐿 the Hermitian fiber metric described in
Proposition 10.5(a), and endow 𝐿 with the corresponding Chern connection. By the

ogy class in 𝐻 𝑝,𝑞 (𝑀; 𝐿) has a 𝜕 𝐿 -harmonic representative. Let 𝛼 ∈ ℋ 𝑝,𝑞 (𝑀; 𝐿)
Hodge–Dolbeault theorem for bundle-valued forms (Thm. 9.35), every cohomol-

be such a form. Because we are using the Chern connection on 𝐿, the exterior co-
variant derivative operator is 𝐷 = 𝐷′ + 𝐷″ where 𝐷″ = 𝜕 𝐿 , and the Dolbeault
Laplacian Δ𝜕𝐿 = 𝜕 𝐿 𝜕 ∗𝐿 + 𝜕 ∗𝐿 𝜕 𝐿 is equal to Δ″ .
The Akizuki–Nakano identity (Thm. 10.4) gives
(10.5) 0 = Δ𝜕𝐿 𝛼 = Δ″ 𝛼 = Δ′ 𝛼 + [Ω, 𝐿∗𝜔 ]𝛼.
On the other hand, our choice of Kähler metric gives
Ω𝛼 = 𝑖Θ𝐿 ∧ 𝛼 = 2𝜋𝐿𝜔 𝛼.
Therefore, (10.5) combined with Lemma 9.39(a) yields
−Δ′ 𝛼 = 2𝜋[𝐿𝜔 , 𝐿∗𝜔 ]𝛼 = 2𝜋(𝑝 + 𝑞 − 𝑛)𝛼.
(Here we are using the fact that 𝐿𝜔 and 𝐿∗𝜔 act only on the differential form part
of a bundle-valued form, so the commutation relation of Lemma 9.39(a) still holds
322 10. The Kodaira Embedding Theorem

for bundle-valued forms.) Taking the global inner product with 𝛼, we find
2𝜋(𝑝 + 𝑞 − 𝑛)‖𝛼‖2 = −(Δ′ 𝛼, 𝛼) = −(𝐷′ 𝐷′∗ 𝛼, 𝛼) − (𝐷′∗ 𝐷′ 𝛼, 𝛼)
= −‖𝐷′∗ 𝛼‖2 − ‖𝐷′ 𝛼‖2 ≤ 0.
When 𝑝 + 𝑞 − 𝑛 > 0, this implies 𝛼 = 0. □

The most useful case of the preceding theorem is the one below; it is the orig-
inal vanishing theorem proved by Kunihiko Kodaira in 1953 [Kod53] and later
generalized as above by Akizuki and Nakano.
Corollary 10.7 (Kodaira Vanishing Theorem). If 𝑀 is a compact complex man-
ifold, 𝐿 → 𝑀 is a positive line bundle, and 𝐾 → 𝑀 is the canonical bundle, then
𝐻 𝑞 (𝑀; 𝒪(𝐾 ⊗ 𝐿)) = 0 for all 𝑞 > 0.

Proof. Note that 𝐾 = Λ𝑛,0 𝑀 (where 𝑛 = dim 𝑀), so 𝒪(𝐾 ⊗ 𝐿) = Ω𝑛 (𝐿). By

the Dolbeault theorem, 𝐻 𝑞 (𝑀; Ω𝑛 (𝐿)) ≅ 𝐻 𝑛,𝑞 (𝑀; 𝐿), and the result follows from
Theorem 10.6. □

In practical applications of the Kodaira vanishing theorem, it is usually more

of a specific line bundle 𝐿, instead of 𝐾 ⊗ 𝐿. The next corollary shows how.

useful to express the conclusion in terms of the cohomology of the sheaf of sections

Corollary 10.8. Suppose 𝑀 is a compact complex manifold and 𝐿 → 𝑀 is a

complex line bundle such that 𝐾 ∗ ⊗ 𝐿 is positive. Then 𝐻 𝑞 (𝑀; 𝒪(𝐿)) = 0 for all
𝑞 > 0.

Proof. This follows from the previous corollary applied to 𝐿′ = 𝐾 ∗ ⊗ 𝐿, noting

that 𝐾 ⊗ 𝐿′ ≅ 𝐿. □

Line Bundles on Blowups

As mentioned at the beginning of this chapter, the strategy for proving the Ko-
daira embedding theorem is to transfer the problem to a blowup, thus converting
sheaves of sections vanishing at a point to sheaves of sections vanishing on a hy-
persurface. The next preliminary result we need is a general fact about lifting a
positive line bundle to a blowup. This is the most technically complicated part of
the proof of the embedding theorem.
Proposition 10.9. Suppose 𝑀 is a compact complex manifold of dimension 𝑛 ≥ 2
and 𝐿 → 𝑀 is a positive line bundle. For any 𝑝 ∈ 𝑀, let 𝜋𝑝 ∶ 𝑀𝑝 → 𝑀 denote
the blowup of 𝑀 at 𝑝, let 𝑆𝑝 = 𝜋𝑝−1 (𝑝) be the exceptional hypersurface, and let
𝐿𝑆𝑝 → 𝑀𝑝 be the line bundle associated with 𝑆𝑝 . There exists an integer 𝑘0 > 0
such that for every integer 𝑘 ≥ 𝑘0 and every point 𝑝 ∈ 𝑀, the following line bundle
on 𝑀𝑝 is positive:
𝜋𝑝∗ 𝐿𝑘 ⊗ 𝐿∗𝑆 .
𝑝
The Kodaira Vanishing Theorem 323

Proof. Give 𝑀 the Kähler metric and 𝐿 the Hermitian fiber metric described in
Proposition 10.5(a), and endow 𝐿 with the corresponding Chern connection, so the
Chern form 2𝜋𝑖 Θ𝐿 is equal to the Kähler form 𝜔. Fix 𝑝 ∈ 𝑀 and let 𝜋𝑝 ∶ 𝑀𝑝 → 𝑀
be the blowup. By Proposition 7.21, the Chern form of 𝜋𝑝∗ 𝐿 with respect to the

𝑖 ∗
pullback metric is equal to
𝜋 Θ = 𝜋𝑝∗ 𝜔.
2𝜋 𝑝 𝐿
Since 𝜋𝑝 restricts to a biholomorphism from 𝑀𝑝 ∖ 𝑆𝑝 to 𝑀 ∖ {𝑝}, and biholomor-
phisms pull positive (1, 1)-forms back to positive (1, 1)-forms, this shows that the
Chern form of 𝜋𝑝∗ 𝐿 is positive on 𝑀𝑝 ∖ 𝑆𝑝 , and thus so is the Chern form of 𝜋𝑝∗ 𝐿𝑘
for all 𝑘 > 0. However, at a point 𝑥 ∈ 𝑆𝑝 , because 𝐷𝜋𝑝 (𝑥)(𝑌 ) = 0 exactly when
𝑌 ∈ 𝑇𝑥 𝑆𝑝 , this form is positive when applied to vectors in 𝑇𝑥 𝑀𝑝 ∖ 𝑇𝑥 𝑆𝑝 , but zero
when restricted to 𝑇𝑥 𝑆𝑝 . We will use the bundle 𝐿∗𝑆 to correct this, at the cost of
introducing some negativity in 𝑀𝑝 ∖ 𝑆𝑝 , which we can eliminate by raising 𝐿 to a
𝑝

Let (𝑣1 , … , 𝑣𝑛 ) be a holomorphic coordinate chart on an open set 𝑊 ⊆ 𝑀

high enough power.

centered at 𝑝, whose image contains a closed ball 𝐵 𝜀 (0) ⊆ ℂ𝑛 . For each 𝑠 ∈ (0, 𝜀],
let 𝑊𝑠 denote the subset of 𝑊 where |𝑣| < 𝑠. Let 𝑊 ̃ = 𝜋𝑝−1 (𝑊𝜀 ) ⊆ 𝑀𝑝 , identified
with the subset 𝑇 (𝜀) of the tautological bundle 𝑇 as in (3.13), so we can write
̃ = {([𝑤], 𝑣) ∶ [𝑤] ∈ ℂℙ𝑛−1 , 𝑣 ∈ [𝑤], |𝑣| < 𝜀}.
𝑊

With this identification, 𝑆𝑝 = {([𝑤], 𝑣) ∶ 𝑣 = 0}, and the restriction of the blow-
down map 𝜋 ∶ 𝑊 ̃ → 𝑊 is given by 𝜋([𝑤], 𝑣) = (𝑣1 , … , 𝑣𝑛 ). There is a holomor-
phic retraction 𝑟 ∶ 𝑊̃ → 𝑆𝑝 given by 𝑟([𝑤], 𝑣) = ([𝑤], 0).
Cover 𝑊 ̃ with open sets 𝑈𝛼 , 𝛼 = 1, … , 𝑛, where 𝑈𝛼 = {([𝑤], 𝑣) ∶ 𝑤𝛼 ≠ 0},
and let 𝑈0 = 𝑀𝑝 ∖ 𝑆𝑝 . We construct a system of local defining functions for 𝑆𝑝 as
follows. On each set 𝑈𝛼 with 𝛼 ≠ 0, we take 𝑓𝛼 = 𝑣𝛼 , which vanishes simply on
𝑈𝛼 ∩ 𝑆𝑝 ; and on 𝑈0 , we take 𝑓0 ≡ 1. Theorem 3.39 shows that {𝑈0 , 𝑈1 , … , 𝑈𝑛 }
is a trivializing cover for 𝐿𝑆𝑝 . The transition function on 𝑈𝛼 ∩ 𝑈𝛽 for nonzero 𝛼, 𝛽
is the holomorphic extension of 𝑣𝛼 /𝑣𝛽 to 𝑈𝛼 ∩ 𝑈𝛽 , which is equal to 𝑤𝛼 /𝑤𝛽 on all
of 𝑈𝛼 ∩ 𝑈𝛽 . On 𝑈𝛼 ∩ 𝑈0 , on the other hand, the transition function is 𝑓𝛼 /𝑓0 = 𝑣𝛼 .
Thus for this trivializing cover, 𝐿𝑆𝑝 has transition functions

𝜏𝛼0 = 𝑣𝛼 on 𝑈𝛼 ∩ 𝑈0 ,
𝑤𝛼
𝜏𝛼𝛽 = on 𝑈𝛼 ∩ 𝑈𝛽 for 𝛼, 𝛽 ∈ {1, … , 𝑛}.
𝑤𝛽

On the other hand, 𝑆𝑝 is biholomorphic to ℂℙ𝑛−1 , and Proposition 3.32 shows

that the tautological bundle 𝐻 ∗ over 𝑆𝑝 is trivial over {𝑈1 ∩ 𝑆𝑝 , … , 𝑈𝑛 ∩ 𝑆𝑝 } with
the restrictions of the same transition functions. Since the retraction 𝑟 ∶ 𝑊 ̃ →
𝑆𝑝 satisfies 𝑟 (𝑤 /𝑤 ) = 𝑤 /𝑤 for each 𝛼 and 𝛽, this shows that the transition
∗ 𝛼 𝛽 𝛼 𝛽
324 10. The Kodaira Embedding Theorem

̃ are the pullbacks by 𝑟 of those for 𝐻 ∗ , so 𝐿𝑆 | ̃ ≅ 𝑟∗ 𝐻 ∗ .

functions for 𝐿𝑆𝑝 over 𝑊 𝑝 𝑊

̃ ≅ 𝑟 𝐻.
Taking duals shows that 𝐿∗𝑆 |𝑊 ∗
𝑝

Because 𝐻 → ℂℙ𝑛−1 is a positive line bundle, it has a Hermitian fiber metric

̃ we can give 𝐿∗ | ̃ the
whose Chern form 2𝜋𝑖 Θ𝐻 is a positive (1, 1)-form, so on 𝑊
pullback metric ⟨⋅, ⋅⟩𝑟∗ 𝐻 , which has Chern form
𝑆𝑝 𝑊

𝑖 𝑖 ∗
Θ𝑟∗ 𝐻 = 𝑟 Θ𝐻 .
2𝜋 2𝜋

tangent to 𝑆𝑝 because 𝑟 restricts to the identity on 𝑆𝑝 .

This form is nonnegative everywhere, and it is positive when restricted to vectors

By Theorem 3.39, there is a global section 𝜎 ∈ 𝒪(𝑀𝑝 , 𝐿𝑆𝑝 ) that vanishes only
on 𝑆𝑝 . Thus the restriction of 𝜎 to 𝑀𝑝 ∖ 𝑆𝑝 is a local holomorphic frame for 𝐿𝑆𝑝
there, and we can define a fiber metric ⟨⋅, ⋅⟩1 on 𝐿𝑆𝑝 |𝑀 by setting ⟨𝜎, 𝜎⟩1 ≡ 1.
𝑝 ∖𝑆𝑝
|𝜎|21
identically zero. Give 𝐿∗𝑆 |𝑀 ∖𝑆 the dual metric, also denoted by ⟨⋅, ⋅⟩1 ; it too has
Because is constant, the curvature of the Chern connection for this metric is
𝑝 𝑝 𝑝

We now create a global fiber metric on 𝐿∗𝑆 by blending these two metrics to-
zero curvature.

gether. Fix once and for all a smooth cutoff function 𝜑 ∶ [0, ∞) → [0, 1] sup-
𝑝

ported in [0, 3𝜀/4) and identically 1 on [0, 𝜀/2], and define a smooth function
𝜌 ∶ 𝑀𝑝 → [0, 1] by 𝜌([𝑤], 𝑣) = 𝜑(|𝑣|) for ([𝑤], 𝑣) ∈ 𝑊 ̃ , extended by zero to
the rest of 𝑀𝑝 . Define a Hermitian fiber metric ⟨⋅, ⋅⟩𝑝 on 𝐿∗𝑆 by
𝑝

⟨⋅, ⋅⟩𝑝 = 𝜌⟨⋅, ⋅⟩𝑟∗ 𝐻 + (1 − 𝜌)⟨⋅, ⋅⟩1 .

The Chern form 2𝜋𝑖 Θ𝑝 associated with ⟨⋅, ⋅⟩𝑝 is nonnegative on 𝜋𝑝−1 (𝑊𝜀/2 ) (where
𝜌 ≡ 1), is strictly positive when restricted to 𝑇 𝑆𝑝 , and is zero on 𝑀𝑝 ∖ 𝜋𝑝−1 (𝑊3𝜀/4 );
but it may have positive and/or negative values on 𝜋𝑝−1 (𝑊3𝜀/4 ∖ 𝑊𝜀/2 ) where 𝜌 is not

Let 𝜎𝑝 ∶ 𝑀 ∖ {𝑝} → 𝑀𝑝 ∖ 𝑆𝑝 be the inverse of the biholomorphism 𝜋𝑝 |𝑀𝑝 ∖𝑆𝑝 ,

constant.

and let Ψ𝑝 = 𝜎𝑝∗ Θ𝑝 , which is a smooth (1, 1)-form on 𝑀 ∖ {𝑝}. It is nonnegative

everywhere on 𝑀 ∖ {𝑝} except perhaps in the compact annulus 𝐾 = 𝑊 3𝜀/4 ∖ 𝑊𝜀/2 .
The set 𝑈 𝑇 𝐾 of unit tangent vectors to 𝐾 is compact [LeeRM, Prop. 2.9], so the
expression 2𝜋𝑖 Ψ𝑝 (𝑋, 𝐽 𝑋) is bounded when applied to unit vectors there. Because
𝑖
Θ (𝑋, 𝐽 𝑋)
2𝜋 𝐿
= 𝜔(𝑋, 𝐽 𝑋) = 𝑔(𝑋, 𝑋) = 1 for every unit vector 𝑋, it follows that
the form
𝑖
𝑘Θ + Ψ𝑝 )
2𝜋 ( 𝐿
(10.6)

is positive on 𝑀 ∖ {𝑝} for 𝑘 sufficiently large.

The Kodaira Vanishing Theorem 325

The tensor product metric on 𝜋𝑝∗ 𝐿𝑘 ⊗ 𝐿∗𝑆 has its Chern form equal to
𝑝

𝑖
𝑘𝜋 ∗ Θ + Θ𝑝 ).
2𝜋 ( 𝑝 𝐿
(10.7)

For 𝑘 sufficiently large, this form is positive on 𝑀𝑝 ∖ 𝑆𝑝 because it is the pullback

of the positive form (10.6); at points of 𝑆𝑝 , both terms are nonnegative, while the
second term is positive when applied to vectors tangent to 𝑆𝑝 , and the first term is

To see that we can choose 𝑘 independently of 𝑝, we start by deriving an explicit

positive when applied to everything else.

formula for the form Ψ𝑝 defined above. Example 7.26 shows that the tautological
bundle 𝐻 ∗ over 𝑆𝑝 ≈ ℂℙ𝑛−1 has a trivializing cover {𝑈1 ∩ 𝑆𝑝 , … , 𝑈𝑛 ∩ 𝑆𝑝 }, and
on 𝑈𝛼 ∩ 𝑆𝑝 we have a local frame 𝑠𝛼 whose norm is

|𝑤|2
|𝑠𝛼 |2𝐻 ∗ = .
|𝑤𝛼 |2

Therefore, the dual frame 𝑠∗𝛼 for 𝐻 has norm

|𝑤𝛼 |2
|𝑠∗𝛼 |2𝐻 = .
|𝑤|2
(10.8)

On each set 𝑈𝛼 , the local frame 𝑠∗𝛼 for 𝐻 pulls back under 𝑟 to a local frame for
𝐿∗𝑆 |𝑈 , which for simplicity we also denote by 𝑠∗𝛼 . The norm of the pullback
𝑝 𝛼

on 𝑈𝛼 because 𝑟∗ (|𝑤𝛼 |2 /|𝑤|2 ) = |𝑤𝛼 |2 /|𝑤|2 . However, on 𝑈𝛼 ∖ 𝑆𝑝 , note that

frame with respect to the pullback metric is given by the same formula (10.8)

|𝑤𝛼 |2 /|𝑤|2 = |𝑣𝛼 |2 /|𝑣|2 , so we can also express the norm there as

|𝑣𝛼 |2
|𝑠∗𝛼 |2𝑟∗ 𝐻 = .
|𝑣|2

On 𝑈𝛼 ∩ 𝑈0 = 𝑈𝛼 ∖ 𝑆𝑝 , on the other hand, Theorem 3.39 shows that the global

section 𝜎 of 𝐿𝑆𝑝 can be expressed as 𝜎 = 𝑓𝛼 𝑠𝛼 = 𝑣𝛼 𝑠𝛼 (no implicit summation),
and therefore the dual frame 𝜎 ∗ for 𝐿∗𝑆 over 𝑈𝛼 ∖ 𝑆𝑝 satisfies 𝜎 ∗ = (1/𝑣𝛼 )𝑠∗𝛼 . Since
|𝜎 ∗ |1 = 1 by design, we have
𝑝

|𝑠∗𝛼 |21 = |𝑣𝛼 |2 .

Thus our blended metric has the local formula

|𝑣𝛼 |2 |𝑣𝛼 |2 (𝜌 + (1 − 𝜌)|𝑣|2 )

|𝑠∗𝛼 |2𝑝 = 𝜌 2 + (1 − 𝜌)|𝑣 | =
𝛼 2
|𝑣| |𝑣|2
326 10. The Kodaira Embedding Theorem

on 𝑈𝛼 ∖ 𝑆𝑝 , and the curvature of its Chern connection is

|𝑣𝛼 |2 (𝜌 + (1 − 𝜌)|𝑣|2 )
Θ𝑝 |𝑈 = 𝜕𝜕 log
𝛼 ∖𝑆𝑝 |𝑣|2
(𝜌 + (1 − 𝜌)|𝑣| )
2
= 𝜕𝜕 log |𝑣𝛼 |2 + log
( |𝑣|2 )
(10.9)

(𝜌 + (1 − 𝜌)|𝑣| )
2
= 𝜕𝜕 log ,
|𝑣|2
where we have used the fact that 𝜕𝜕 log |𝑣𝛼 |2 = 0, which can be seen by writing
log |𝑣𝛼 |2 = log 𝑣𝛼 + log 𝑣𝛼 in a neighborhood of each point where a branch of the

independent of 𝛼, so it actually holds on all of 𝑊 ̃ ∖ 𝑆𝑝 . Since the restriction of 𝜎𝑝

complex logarithm is defined. The notable thing about this formula is that it is

to 𝑊𝜀 ∖ {𝑝} is given by 𝜎𝑝 (𝑣) = ([𝑣], 𝑣), it follows that the pullback Ψ𝑝 = 𝜎𝑝∗ Θ𝑝 is
given by the same formula (10.9) on 𝑊𝜀 ∖ {𝑝}.
Now let 𝑞 = (𝑞 1 , … , 𝑞 𝑛 ) be any other point in the open set 𝑊𝜀/4 . We can carry
out the same construction of a fiber metric on 𝐿∗𝑆 using coordinates (𝑣1̃ , … , 𝑣𝑛̃ ),
𝑞
where 𝑣𝑗̃ = 𝑣𝑗 − 𝑞 𝑗 . Defining 𝜎𝑞 ∶ 𝑀 ∖ {𝑞} → 𝑀𝑞 and Ψ𝑞 = 𝜎𝑞∗ Θ𝑞 as above, we
find that that construction results in
̃ + (1 − 𝜑(|𝑣|))|
(𝜑(|𝑣|) ̃ 𝑣|̃ ) 2
Ψ𝑞 = 𝜕𝜕 log
|𝑣|̃ 2
(𝜑(|𝑣 − 𝑞|) + (1 − 𝜑(|𝑣 − 𝑞|))|𝑣 − 𝑞| )
2
= 𝜕𝜕 log
|𝑣 − 𝑞|2
on 𝑊 𝜀 ∖ {𝑞}. For any such 𝑞, the form Ψ𝑞 is nonnegative except possibly in the
compact annulus 𝐾 ′ = 𝑊 𝜀 ∖ 𝑊𝜀/4 , and the expression 2𝜋𝑖 Ψ𝑞 |𝑥 (𝑋, 𝐽 𝑋) depends
continuously on 𝑞 ∈ 𝑊𝜀/4 and (𝑥, 𝑋) ∈ 𝑈 𝑇 𝐾 ′ . Since 𝑊 𝜀/8 × 𝑈 𝑇 𝐾 ′ is compact,
there exists an integer 𝑘𝑝 such that 2𝜋𝑖 (𝑘Θ𝐿 + Ψ𝑞 ) is positive for all 𝑘 ≥ 𝑘𝑝 and
all 𝑞 in the neighborhood 𝑊𝜀/8 of 𝑝, which implies as above that 𝜋𝑞∗ 𝐿𝑘 ⊗ 𝐿∗𝑆 is
positive for all such 𝑘 and 𝑞. Since 𝑀 is compact, we can cover it with finitely
𝑞

many such neighborhoods and let 𝑘0 be the largest such integer, thus completing
the proof. □

For proving that 𝒪(𝑀; 𝐿𝑘 ) separates points, we will also need an adaptation of
the previous result for blowups at two points.
Proposition 10.10. With 𝑀 and 𝐿 as in the hypothesis of Proposition 10.9, let 𝑘0 be
the integer guaranteed by that proposition. For a pair of distinct points 𝑝, 𝑞 ∈ 𝑀,
let 𝜋𝑝𝑞 ∶ 𝑀𝑝𝑞 → 𝑀 be the blowup of 𝑀 at 𝑝 and 𝑞, and let 𝑆𝑝 = 𝜋𝑝𝑞 −1
({𝑝}) and
𝑆𝑞 = 𝜋𝑝𝑞 −1
({𝑞}). For any 𝑘 ≥ 𝑘0 and any distinct 𝑝, 𝑞 ∈ 𝑀, the following line
The Kodaira Vanishing Theorem 327

bundle on 𝑀𝑝𝑞 is positive:

𝜋𝑝𝑞 𝐿 ⊗ 𝐿∗𝑆 ⊗ 𝐿∗𝑆 .

∗ 2𝑘
𝑝 𝑞

Proof. Let 𝑝, 𝑞 ∈ 𝑀 be an arbitrary pair of distinct points, and 𝑘 ≥ 𝑘0 . If we let

̃ 𝑝 → 𝑀 and 𝜋̃𝑞 ∶ 𝑀
𝜋̃𝑝 ∶ 𝑀 ̃ 𝑞 → 𝑀 be the blowups of 𝑀 at 𝑝 and 𝑞, respectively,
we have a commutative diagram of blowdown maps:
𝑀𝑝𝑞
𝜋𝑞 ④④ ❈❈ 𝜋
❈❈ 𝑝

̃𝑝 𝜋𝑝𝑞 ̃𝑞
④

𝑀 𝑀
④ ❈❈
④④ ❈❈
}④
④ !

𝜋̃𝑝 ❊❊❊"  |②②②② 𝜋̃𝑞

❊❊
❊❊ ②②
②②

𝑀.
❊

With 𝑆̃𝑝 = 𝜋̃−1 ̃

𝑝 ({𝑝}) and 𝑆 𝑞 = 𝜋 ̃−1
𝑞 ({𝑞}), Proposition 10.9 shows that both of
the bundles 𝜋̃∗𝑝 𝐿𝑘 ⊗ 𝐿∗𝑆̃ → 𝑀 ̃ 𝑝 and 𝜋̃∗𝑞 𝐿𝑘 ⊗ 𝐿∗̃ → 𝑀 ̃ 𝑞 are positive. Because
𝑆𝑞
𝜋𝑞 ∶ 𝑀𝑝𝑞 → 𝑀 ̃ 𝑝 is a biholomorphism from a neighborhood of 𝑆𝑝 to a neighbor-
𝑝

̃
hood of 𝑆 𝑝 , it follows from Problem 3-12 that 𝜋𝑞∗ 𝐿𝑆̃𝑝 ≅ 𝐿𝑆𝑝 . Thus the bundle
𝜋𝑝𝑞 𝐿 ⊗ 𝐿∗𝑆 ≅ 𝜋𝑞∗ (𝜋̃∗𝑝 𝐿𝑘 ⊗ 𝐿∗𝑆̃ ) has a Chern form that is positive on 𝑀𝑝𝑞 ∖ 𝑆𝑞 ,
∗ 𝑘

where 𝜋𝑞 is a biholomorphism onto its image; at points of 𝑆𝑞 , it is merely nonneg-

𝑝 𝑝

ative. Similarly, 𝜋𝑝𝑞

∗ 𝑘
𝐿 ⊗ 𝐿∗𝑆 has a Chern form that is positive on 𝑀𝑝𝑞 ∖ 𝑆𝑝 and
nonnegative at points of 𝑆𝑝 . Therefore, the tensor product of these two bundles,
𝑞

namely

(𝜋𝑝𝑞 𝐿 ⊗ 𝐿𝑆𝑝 ) ⊗ (𝜋𝑝𝑞 𝐿 ⊗ 𝐿𝑆𝑞 ) ≅ 𝜋𝑝𝑞 𝐿 ⊗ 𝐿𝑆𝑝 ⊗ 𝐿𝑆𝑞 ,

∗ 𝑘 ∗ ∗ 𝑘 ∗ ∗ 2𝑘 ∗ ∗

has a Chern form that is the sum of the two, which is positive everywhere. □

Because the Kodaira vanishing theorem involves the canonical bundle, it will
also be important to have explicit information about the canonical bundle of a
blowup.
Proposition 10.11. Suppose 𝑀 is a complex 𝑛-manifold and 𝜋 ∶ 𝑀 ̃ → 𝑀 is the
blowup of 𝑀 at finitely many points 𝑝1 , … , 𝑝𝑚 ∈ 𝑀. Let 𝐾 denote the canonical
bundle of 𝑀, 𝐾 ̃ the canonical bundle of 𝑀,̃ and for each 𝑖, 𝐿𝑆 the line bundle
associated with the exceptional hypersurface 𝑆𝑖 = 𝜋 −1 ({𝑝𝑖 }). Then
𝑖

̃ ≅ 𝜋 ∗ 𝐾 ⊗ 𝐿𝑛−1 ⊗ ⋯ ⊗ 𝐿𝑛−1 .
𝐾
(10.10) 𝑆 1 𝑆 𝑚

Proof. For simplicity, we will prove the proposition for the blowup at one point.
The generalization of the proof to multiple points is more complicated notationally
but not conceptually, so it is left to the reader.
328 10. The Kodaira Embedding Theorem

Thus let 𝑀̃ be the blowup of 𝑀 at 𝑝 ∈ 𝑀, and let 𝐿𝑆 be the line bundle

associated with the exceptional hypersurface 𝑆 = 𝜋 −1 ({𝑝}). We will prove that

̃ ≅ 𝜋 ∗ 𝐾 ⊗ 𝐿𝑛−1
𝐾
(10.11) 𝑆

by showing that the bundles on both sides of the equality have a trivializing cover

We begin by constructing a trivializing cover of 𝑀 for 𝐾. Choose a holo-

with the same transition functions.

morphic coordinate chart (𝑊0 , (𝑣1 , … , 𝑣𝑛 )) centered at 𝑝, whose image is a ball

𝐵𝜀 (0) ⊆ ℂ𝑛 ; and let {𝑊𝑐 }𝑐∈𝐶 be a cover of 𝑀 ∖ {𝑝} by holomorphic coordinate
domains, with coordinates (𝜁(𝑐)
1
, … , 𝜁(𝑐)
𝑛
) on 𝑊𝑐 . In each such open set there is a
nonvanishing holomorphic section of 𝐾, namely 𝑠𝑐 = 𝑑𝜁(𝑐) 1
∧ ⋯ ∧ 𝑑𝜁(𝑐)
𝑛
in 𝑊𝑐 and
𝑠0 = 𝑑𝑣1 ∧ ⋯ ∧ 𝑑𝑣𝑛 in 𝑊0 . The transition functions for this cover are given by

𝜕𝜁(𝑑)
𝑗

𝜏𝑐𝑑
𝐾
= det on 𝑊𝑐 ∩ 𝑊𝑑 ,
( 𝜕𝜁 𝑘 )
(𝑐)

𝜕𝜁(𝑑)
𝑗

𝜏0𝑑
𝐾
= on 𝑊0 ∩ 𝑊𝑑 .
( 𝜕𝑣𝑘 )
det

Next we construct a trivializing cover of 𝑀 ̃ for 𝐾. ̃ The map 𝜋 restricts to a

biholomorphism from 𝑀 ̃ ∖ 𝑆 to 𝑀 ∖ {𝑝}, so we can define 𝑊 ̃ 𝑐 = 𝜋 −1 (𝑊𝑐 ) for
𝑐 ∈ 𝐶, and on each such set there is a holomorphic local frame 𝑠𝑐̃ = 𝜋 ∗ 𝑠𝑐 for 𝐾. ̃
̃ 0 = 𝜋 (𝑊0 ) ⊆ 𝑀,
We let 𝑊 −1 ̃ identified with 𝑇 (𝜀) as in the proof of Proposition
̃ 0 with open sets 𝑈𝛼 = {([𝑤], 𝑣) ∶ 𝑤𝛼 ≠ 0}. On each
10.9. As before, we cover 𝑊
such set there are holomorphic coordinates (𝑧1(𝛼) , … , 𝑧𝑛(𝛼) ) defined by

⎧ 𝑤 , 𝑗 ≠ 𝛼,
𝑗
⎪ 𝛼
𝑧(𝛼) = ⎨𝑤
𝑗

⎪𝑣𝛼 , 𝑗 = 𝛼,
⎩
and the inverse of the coordinate map is given by

([𝑤], 𝑣) = (([𝑧1(𝛼) , … , 1 … , 𝑧𝑛(𝛼) ]), 𝑧𝛼(𝛼) (𝑧1(𝛼) , … , 1, … … , 𝑧𝑛(𝛼) )),

with the 1’s in the 𝛼th positions. (The motivation for these coordinates comes from
the description of 𝑇 as the tautological bundle over ℂℙ𝑛−1 : the 𝛼th coordinate is
a fiber coordinate for 𝑇 , and the remaining coordinates are affine coordinates for
ℂℙ𝑛−1 .)
Over each 𝑈𝛼 , there is a holomorphic local frame for 𝐾 ̃ given by

𝑠𝛼̃ = 𝑑𝑧1(𝛼) ∧ ⋯ ∧ 𝑑𝑧𝑛(𝛼) .

The Kodaira Vanishing Theorem 329

̃ 0 . On the complement of 𝑆 in 𝑈𝛼 , we
in terms of the functions (𝑣1 , … , 𝑣𝑛 ) on 𝑊
To compute the transition functions for these frames, it is easiest to express them

have 𝑤 /𝑤 = 𝑣 /𝑣 , so
𝑗 𝛼 𝑗 𝛼

𝑣1 𝑣𝑛
𝑠𝛼̃ = 𝑑 ( 𝛼 ) ∧ ⋯ ∧ 𝑑𝑣𝛼 ∧ ⋯ ∧ 𝑑 ( 𝛼 )
𝑣 𝑣
1 𝑛−1 1
= ( 𝛼 ) 𝑑𝑣 ∧ ⋯ ∧ 𝑑𝑣𝛼 ∧ ⋯ ∧ 𝑑𝑣𝑛
𝑣
1 𝑛−1
= ( 𝛼 ) 𝜋 ∗ 𝑠0 ,
𝑣

derivatives of 𝑣𝛼 cancel because 𝑑𝑣𝛼 ∧ 𝑑𝑣𝛼 = 0. Therefore 𝑠𝛽̃ = (𝑣𝛼 /𝑣𝛽 )𝑛−1 𝑠𝛼̃ =
where the second equality follows from the fact that the remaining terms involving

(𝑤𝛼 /𝑤𝛽 )𝑛−1 𝑠𝛼̃ on (𝑈𝛼 ∩ 𝑈𝛽 ) ∖ 𝑆, and by continuity 𝑠𝛽̃ = (𝑤𝛼 /𝑤𝛽 )𝑛−1 𝑠𝛼̃ on all of
̃ 𝑐 ∶ 𝑐 ∈ 𝐶} is a trivializing
𝑈𝛼 ∩ 𝑈𝛽 . It follows that the collection {𝑈1 , … , 𝑈𝑛 , 𝑊
̃ ̃
cover of 𝑀 for 𝐾, with transition functions

𝜏𝑐𝑑
̃
𝐾
= 𝜋 ∗ 𝜏𝑐𝑑
𝐾 ̃𝑐 ∩ 𝑊
on 𝑊 ̃𝑑,
𝜏𝛼𝑑
𝐾̃
= (𝑣𝛼 )𝑛−1 𝜋 ∗ 𝜏0𝑑
𝐾 ̃𝑑,
on 𝑈𝛼 ∩ 𝑊
𝑤𝛼 𝑛−1
𝜏𝛼𝛽 = ( 𝛽) on 𝑈𝛼 ∩ 𝑈𝛽 .
𝐾̃
𝑤

This same cover is a trivalizing cover for 𝜋 ∗ 𝐾. Since 𝐾 is trivial over 𝑊0 , it

follows that 𝜋 ∗ 𝐾 is trivial over each 𝑈𝛼 , and it has transition functions

𝜏𝑐𝑑
𝜋 𝐾
= 𝜋 ∗ 𝜏𝑐𝑑
𝐾 ̃𝑐 ∩ 𝑊
on 𝑊 ̃𝑑,
∗

𝜏0𝑑
𝜋 𝐾
= 𝜋 ∗ 𝜏0𝑑
𝐾 ̃𝑑,
on 𝑈𝛼 ∩ 𝑊
∗

𝜏𝛼𝛽
𝜋 𝐾
=1 on 𝑈𝛼 ∩ 𝑈𝛽 .
∗

Finally, the proof of Proposition 10.9 shows that there are trivializations for 𝐿𝑆
over the same open cover, with transition functions

𝜏𝑐𝑑𝑆 = 1
𝐿
̃𝑐 ∩ 𝑊
on 𝑊 ̃𝑑,

𝜏𝛼𝑑𝑆 = 𝑣𝛼
𝐿
̃𝑑,
on 𝑈𝛼 ∩ 𝑊
𝑤𝛼
𝜏𝛼𝛽𝑆 = on 𝑈𝛼 ∩ 𝑈𝛽 .
𝐿
𝑤𝛽
̃ satisfy
Putting these results together, we see that the transition functions for 𝐾

𝜏⋅⋅ = (𝜏⋅⋅𝑆 ) 𝜏⋅⋅

̃
𝐾 𝐿 𝑛−1 𝜋 ∗ 𝐾

for each pair of indices, which proves the result. □

330 10. The Kodaira Embedding Theorem

Proof of the Embedding Theorem

Here is the first version of the embedding theorem.

𝑀 is a compact complex manifold. A holomorphic line bundle 𝐿 → 𝑀 is ample

Theorem 10.12 (Kodaira Embedding Theorem, Line Bundle Version). Suppose

if and only if it is positive. Thus 𝑀 is projective if and only if it admits a positive

holomorphic line bundle.

Proof. For the 1-dimensional case, see Problem 10-3. We assume from now on
that 𝑀 has dimension 𝑛 ≥ 2.
Let 𝐿 → 𝑀 be a holomorphic line bundle. Suppose first that 𝐿 is ample,
meaning that some positive tensor power 𝐿𝑘 is very ample. Thus by Theorem 3.43,
its associated map 𝐹 ∶ 𝑀 → ℂℙ𝑁 is an embedding for some 𝑁. Proposition 3.45
shows that 𝐿𝑘 ≅ 𝐹 ∗ 𝐻, so 𝐿𝑘 is positive by Proposition 10.5.
Given a Hermitian fiber metric ⟨⋅, ⋅⟩𝐿𝑘 on 𝐿𝑘 whose Chern form is positive,
there is a unique Hermitian fiber metric ⟨⋅, ⋅⟩𝐿 on 𝐿 such that ⟨⋅, ⋅⟩𝐿𝑘 is the tensor
product metric of that on 𝐿, namely
|𝑣|𝐿 = (|𝑣 ⊗ ⋯ ⊗ 𝑣|𝐿𝑘 ) .
1/𝑘

The Chern form of 𝐿 with respect to this metric is

𝑖 𝑖 1
Θ = Θ 𝑘,
2𝜋 𝐿 2𝜋 𝑘 𝐿
which is also positive. This proves that 𝐿 is a positive line bundle.
Conversely, suppose 𝐿 is positive. Using Proposition 10.5(a), give 𝑀 a Kähler
metric and 𝐿 a Hermitian fiber metric whose Chern form 2𝜋𝑖 Θ𝐿 is equal to the
Kähler form 𝜔. Choose a fiber metric on the anticanonical bundle 𝐾 ∗ and let Θ𝐾 ∗
be the curvature of its associated Chern connection. Because 2𝜋𝑖 Θ𝐿 is positive,
there is some 𝑘1 > 0 such that for every 𝑘′ ≥ 𝑘1 , the (1, 1)-form
𝑖
(Θ ∗ + 𝑘′ Θ𝐿 )
2𝜋 𝐾
is positive on 𝑀, and thus 𝐾 ∗ ⊗ 𝐿𝑘 is a positive line bundle.
′

Suppose 𝑘 is any positive integer such that 𝑘 ≥ 2𝑛𝑘0 +𝑘1 , where 𝑘0 is the integer
guaranteed by Proposition 10.9 and 𝑘1 is defined in the preceding paragraph.
We will show first that 𝒪(𝑀; 𝐿𝑘 ) separates points. As explained at the begin-
ning of this chapter, this is equivalent to showing that the evaluation map
𝑒𝑀 ∶ 𝒪(𝑀; 𝐿𝑘 ) → 𝐿𝑘𝑝 ⊕ 𝐿𝑘𝑞
is surjective for each pair of distinct points 𝑝, 𝑞 ∈ 𝑀. Let 𝑝, 𝑞 ∈ 𝑀 be arbitrary
distinct points and consider the following short exact sheaf sequence on 𝑀:
𝑒
(10.12) 0 → ℐ{𝑝,𝑞} (𝐿𝑘 ) ↪ 𝒪(𝐿𝑘 ) → (𝐿𝑝𝑘 )𝑝 ⊕ (𝐿𝑘𝑞 )𝑞 → 0.
Proof of the Embedding Theorem 331

𝒪(𝐿𝑘 )/ℐ{𝑝,𝑞} (𝐿𝑘 ), so our problem is equivalent to showing the surjectivity of the
By Proposition 5.22, the sheaf on the right is isomorphic to the quotient sheaf

global section map Π𝑀 in the following sequence:

Π𝑀
(10.13) 0 → ℐ{𝑝,𝑞} (𝑀; 𝐿𝑘 ) ↪ 𝒪(𝑀; 𝐿𝑘 ) ⟶ Γ(𝒪(𝐿𝑘 )/ℐ{𝑝,𝑞} (𝐿𝑘 )),

Π𝑀 is the global section map associated with the canonical sheaf homomorphism
where the right-hand group is the group of global sections of the quotient sheaf and

Π given by Proposition 5.22. (Bear in mind that this group of global sections is ob-

group 𝒪(𝑀; 𝐿𝑘 )/ℐ{𝑝,𝑞} (𝑀; 𝐿𝑘 ).)

tained from the sheafification functor, and is generally not the same as the quotient

Let 𝜋 ∶ 𝑀𝑝𝑞 → 𝑀 be the blowup of 𝑀 at 𝑝 and 𝑞. Let 𝑆𝑝 = 𝜋 −1 (𝑝), 𝑆𝑞 =

̃ = 𝜋 ∗ 𝐿. There is an analogous short exact sheaf
𝜋 −1 (𝑞), 𝑆 = 𝑆𝑝 ∪ 𝑆𝑞 , and 𝐿
sequence on 𝑀𝑝𝑞 :
̃
̃ 𝑘 ) ↪ 𝒪(𝐿
̃ 𝑘 ) → 𝒪(𝐿
̃ 𝑘 )/ℐ𝑆 (𝐿
̃ 𝑘 )) → 0.
Π
(10.14) 0 → ℐ 𝑆 (𝐿
We can combine the global section sequence of this sheaf sequence with (10.13) to
obtain the following commutative diagram:
̃𝑀
Π
0 𝑆 𝑝𝑞
̃𝑘) 
/ ℐ (𝑀 ; 𝐿 ̃𝑘)
/ 𝒪(𝑀 ; 𝐿
𝑝𝑞
𝑝𝑞
̃ 𝑘 )/ℐ𝑆 (𝐿
/ Γ(𝒪(𝐿 ̃ 𝑘 ))


𝜋0∗ 𝜋∗ 𝜋1∗
O O O
(10.15)
Π𝑀
0 /ℐ
{𝑝,𝑞} (𝑀; 𝐿 )
𝑘  / 𝒪(𝑀; 𝐿𝑘 ) / Γ(𝒪(𝐿𝑘 )/ℐ
{𝑝,𝑞} (𝐿 )),
𝑘


where 𝜋 ∗ is the pullback operator on sections (see Prop. 3.10), 𝜋0∗ is the restriction
of 𝜋 ∗ , and 𝜋1∗ is obtained from 𝜋 ∗ by passing to the quotient: specifically, given
a global section 𝜎 of the quotient sheaf 𝒪(𝐿𝑘 )/ℐ{𝑝,𝑞} (𝐿𝑘 ), for each 𝑥̃ ∈ 𝑀𝑝𝑞 , we
choose a representative 𝑠 ∈ 𝒪(𝑈 ; 𝐿𝑘 )/ℐ{𝑝,𝑞} (𝑈 ; 𝐿𝑘 ) for the germ 𝜎(𝜋(𝑥)), and
̃ 𝑘 )/ℐ𝑆 (𝜋 −1 (𝑈 ); 𝐿
define (𝜋1∗ 𝜎)(𝑥)̃ to be the germ of 𝜋 ∗ 𝑠 in 𝒪(𝜋 −1 (𝑈 ); 𝐿 ̃ 𝑘 ). Because
𝜋 𝑠 vanishes on 𝑆 whenever 𝑠 vanishes at 𝑝 and 𝑞, this is well defined.
∗

We will show that 𝜋 ∗ is surjective and 𝜋1∗ is injective. For 𝜋∗ , suppose 𝜎

̃ ∈
̃
𝒪(𝑀𝑝𝑞 ; 𝐿 ) is arbitrary. Because 𝜋 restricts to a biholomorphism from 𝑀𝑝𝑞 ∖ 𝑆
𝑘

to 𝑀 ∖ {𝑝, 𝑞}, we can define a holomorphic section 𝜎 ∈ 𝒪(𝑀 ∖ {𝑝, 𝑞}; 𝐿𝑘 ) by

pulling back 𝜎 ̃ by the inverse of 𝜋|𝑀𝑝𝑞 ∖𝑆 . Choosing holomorphic coordinates for
𝑀 and a holomorphic local frame 𝑠 for 𝐿 in a neighborhood 𝑈 of 𝑝, we can write the
restriction of 𝜎 to 𝑈 ∖{𝑝} as 𝜎 = 𝑓 𝑠𝑘 for some holomorphic function 𝑓 ∶ 𝑈 ∖{𝑝} →
ℂ. Hartogs’s extension theorem shows that this function extends holomorphically
to all of 𝑈 . A similar argument works in a neighborhood of 𝑞, so we obtain a
globally defined section 𝜎 ∈ 𝒪(𝑀; 𝐿𝑘 ) satisfying 𝜋 ∗ 𝜎 = 𝜎 ̃ on 𝑀𝑝𝑞 ∖ 𝑆, and by
continuity on all of 𝑀𝑝𝑞 . Thus 𝜋 ∗ is surjective.
332 10. The Kodaira Embedding Theorem

To see that 𝜋1∗ is injective, suppose 𝜎 is a global section of the quotient sheaf
𝒪(𝐿𝑘 )/ℐ{𝑝,𝑞} (𝐿𝑘 ) satisfying 𝜋1∗ 𝜎 = 0. Given 𝑥 ∈ 𝑀 and a representative sec-
tion 𝑠 ∈ 𝒪(𝑈 ; 𝐿𝑘 ) for the germ 𝜎(𝑥), let 𝑥̃ be any point in 𝜋 −1 (𝑥). The fact that
̃; 𝐿
(𝜋1∗ 𝜎)(𝑥)̃ = 0 means that the pullback 𝜋 ∗ 𝑠 lies in ℐ𝑆 (𝑈 ̃ 𝑘 ) on some neighborhood
𝑈̃ of 𝑥.̃ Since 𝜋 𝑠(𝑥)̃ = (𝑥,̃ 𝑠(𝑥)), this implies that 𝑠(𝑥) = 0 when 𝑥 is equal to 𝑝
∗

or 𝑞, so 𝑠 represents zero in the quotient space 𝒪(𝑈 ; 𝐿𝑘 )/ℐ{𝑝,𝑞} (𝑈 ; 𝐿𝑘 ) and thus

𝜎(𝑥) = 0.

to show that the map Π̃ 𝑀 in (10.15) is surjective. From the long exact cohomol-
Now comes the heart of the proof. We will use the Kodaira vanishing theorem
𝑝𝑞

̃ 𝑘 )) = 0. By Proposition 5.16, the sheaf ℐ𝑆 (𝐿

the case provided 𝐻 1 (𝑀𝑝𝑞 ; ℐ𝑆 (𝐿 ̃𝑘)
ogy sequence associated with the sheaf sequence (10.14), we see that this will be

̃ 𝑘 ⊗ 𝐿∗ ), which in turn is isomorphic to 𝒪(𝐿

is isomorphic to 𝒪(𝐿 ̃ 𝑘 ⊗ 𝐿∗ ⊗ 𝐿∗ )
𝑆 𝑆𝑝 𝑆𝑞

̃ 𝑘 ⊗ 𝐿∗ ⊗ 𝐿∗ )) = 0 provided that 𝐾
𝐻 1 (𝑀𝑝𝑞 ; 𝒪(𝐿 ̃∗ ⊗ 𝐿 ̃ 𝑘 ⊗ 𝐿∗ ⊗ 𝐿∗ is a
by Problem 3-11. Corollary 10.8 to the Kodaira vanishing theorem shows that
𝑆 𝑆 𝑆 𝑆
̃ is the canonical bundle of 𝑀𝑝𝑞 . But Proposition
𝑝 𝑞 𝑝 𝑞
positive line bundle, where 𝐾

̃ ≅ 𝜋 ∗ 𝐾 ⊗ 𝐿𝑛−1 ⊗ 𝐿𝑛−1 ,
𝐾
10.11 shows that
𝑆𝑝 𝑆𝑞

from which it follows that

̃∗ ⊗ 𝐿
𝐾 ̃ 𝑘 ⊗ 𝐿∗ ⊗ 𝐿∗ ≅ 𝜋 ∗ (𝐾 ∗ ⊗ 𝐿𝑘 ) ⊗ (𝐿∗ )𝑛 ⊗ (𝐿∗ )𝑛 .
(10.16) 𝑆 𝑝𝑆 𝑞 𝑆 𝑆
𝑝 𝑞

Because 𝑘 ≥ 2𝑛𝑘0 + 𝑘1 , we can write 𝑘 = 2𝑛𝑘0 + 𝑘′ with 𝑘′ ≥ 𝑘1 , and rewrite the

tensor product on the right-hand side of (10.16) as

𝜋 ∗ (𝐾 ∗ ⊗ 𝐿𝑘 ) ⊗ (𝜋 ∗ 𝐿2𝑘0 ⊗ 𝐿∗𝑆 ⊗ 𝐿∗𝑆 ) .

′ 𝑛
𝑝 𝑞

The first bundle in this tensor product is the pullback of a positive bundle, which is

product bundle is positive, which completes the proof that Π ̃ 𝑀 is surjective.

therefore nonnegative; the second is positive by Proposition 10.10. Thus the tensor
𝑝𝑞
Using this result, we can show that the map Π𝑀 in (10.14) is also surjective.
Let 𝜎 ∈ Γ(𝒪(𝐿𝑘 )/ℐ{𝑝,𝑞} (𝐿𝑘 )) be arbitrary. The surjectivity of Π ̃ 𝑀 shows that
̃ 𝑀 (𝜏)̃ = 𝜋 ∗ (𝜎). Since 𝜋 ∗ is surjective,
̃ 𝑘 ) such that Π
𝑝𝑞
there exists 𝜏 ̃ ∈ 𝒪(𝑀𝑝𝑞 ; 𝐿 𝑝𝑞 1
there exists 𝜏 ∈ 𝒪(𝑀; 𝐿𝑘 ) such that 𝜋 ∗ (𝜏) = 𝜏,̃ and then commutativity of (10.15)

̃ 𝑀 𝜋 ∗ (𝜏) = Π
𝜋1∗ Π𝑀 (𝜏) = Π ̃ 𝑀 (𝜏)̃ = 𝜋 ∗ 𝜎,
implies
𝑝𝑞 𝑝𝑞 1
and the fact that 𝜋1∗ is injective means that Π𝑀 (𝜏) = 𝜎. This completes the proof
that 𝒪(𝑀; 𝐿𝑘 ) separates points.
Next we address the question of separating directions. Let 𝑝 ∈ 𝑀 be arbitrary
and let 𝑠 be a holomorphic local frame for 𝐿 on a neighborhood of 𝑝. Using the
result of Exercise 10.1, we need to show that the map 𝛿𝑀 ∶ ℐ{𝑝} (𝑀; 𝐿𝑘 ) → Λ𝑝 𝑀
1,0
Proof of the Embedding Theorem 333

given by 𝛿𝑀 (𝑓 𝑠𝑘 ) = 𝑑𝑓𝑝 is surjective. There is a short exact sheaf sequence

𝛿
0 → ℐ{𝑝} (𝐿𝑘 ) ↪ ℐ{𝑝} (𝐿𝑘 ) → (Λ𝑝 𝑀)𝑝 → 0,
2 1,0
(10.17)

and the sheaf on the right is isomorphic to the quotient sheaf ℐ{𝑝} (𝐿𝑘 )/ℐ{𝑝}
2
(𝐿𝑘 ),
so we need to show the surjectivity of Π𝑀 in the following sequence:
Π𝑀
(10.18) 0 → ℐ{𝑝}
2
(𝑀; 𝐿𝑘 ) ↪ ℐ{𝑝} (𝑀; 𝐿𝑘 ) ⟶ Γ(ℐ{𝑝} (𝐿𝑘 )/ℐ{𝑝}
2
(𝐿𝑘 )).

̃ = 𝜋 ∗ 𝐿. We
Let 𝜋 ∶ 𝑀𝑝 → 𝑀 be the blowup of 𝑀 at 𝑝, 𝑆 = 𝜋 −1 (𝑝), and 𝐿
have a short exact sheaf sequence on 𝑀𝑝 :
̃
̃ 𝑘 ) ↪ ℐ 𝑆 (𝐿
̃ 𝑘 ) → ℐ𝑆 (𝐿
̃ 𝑘 )/ℐ 2 (𝐿
̃ 𝑘 )) → 0,
Π
0 → ℐ𝑆2 (𝐿 𝑆

and a commutative diagram of global section maps:

̃𝑀
Π
0 ̃𝑘) 
/ ℐ 2 (𝑀 ; 𝐿 ̃𝑘)
/ ℐ (𝑀 ; 𝐿 𝑝
̃ 𝑘 )/ℐ 2 (𝐿
/ Γ(ℐ (𝐿 ̃ 𝑘 ))
𝑆 O𝑝 𝑆
O𝑝 𝑆
O 𝑆


(10.19) 𝜋0∗ 𝜋∗ 𝜋1∗

Π
0 / ℐ 2 (𝑀; 𝐿𝑘 )  
{𝑝}
/ ℐ (𝑀; 𝐿𝑘 ) 𝑀/ Γ(ℐ (𝐿𝑘 )/ℐ 2 (𝐿𝑘 )).
{𝑝} {𝑝} {𝑝}

As before, 𝜋 ∗ is surjective. We can also show that 𝜋1∗ is injective, but it takes a little
more work this time. Suppose 𝜎 ∈ Γ(ℐ{𝑝} (𝐿𝑘 )/ℐ{𝑝} 2
(𝐿𝑘 )) with 𝜋1∗ 𝜎 = 0. Since
the stalks of ℐ{𝑝} (𝐿𝑘 )/ℐ{𝑝}
2
(𝐿𝑘 ) are zero at all points of 𝑀 other than 𝑝, we need
only show 𝜎(𝑝) = 0. Choose holomorphic coordinates (𝑣1 , … , 𝑣𝑛 ) on a domain
𝑊0 centered at 𝑝 with image 𝐵𝜀 (0) ⊆ ℂ𝑛 , and identify 𝑊 ̃ = 𝜋 −1 (𝑊0 ) ⊆ 𝑀𝑝
with {([𝑤], 𝑣) ∈ ℂℙ𝑛−1 × 𝐵𝜀 (0) ∶ 𝑣 ∈ [𝑤]} as in the proof of Proposition 10.9.
After shrinking 𝑊0 if necessary, we can choose a section 𝑠0 ∈ ℐ{𝑝} (𝑊0 ; 𝐿𝑘 ) whose
image in the quotient space ℐ{𝑝} (𝑊0 ; 𝐿𝑘 )/ℐ{𝑝}
2
(𝑊0 ; 𝐿𝑘 ) is a representative of the
germ 𝜎(𝑝). Because 𝑠0 vanishes at 𝑝, we can write its coordinate representation in
𝑊0 as
𝑠0 (𝑣) = 𝑣𝑗 𝑓𝑗 (𝑣)𝑠(𝑣) ⊗ ⋯ ⊗ 𝑠(𝑣),
∑
𝑗

for some holomorphic functions 𝑓1 , … , 𝑓𝑛 , where 𝑠 is our chosen local frame for
𝐿. The hypothesis implies (after shrinking 𝑊0 further if necessary) that 𝜋 ∗ 𝑠0 ∈
̃; 𝐿
ℐ𝑆2 (𝑊 ̃ 𝑘 ). To show that 𝑠0 ∈ ℐ 2 (𝑊0 ; 𝐿𝑘 ) and therefore 𝜎(𝑝) = 0, we need
{𝑝}
to show that 𝑓𝑗 (𝑝) = 0 for each 𝑗. Let 𝑗 ∈ {1, … , 𝑛} be arbitrary, and define a
̃ by
function 𝜆𝑗 ∶ 𝐷𝜀 (0) → 𝑊

𝜆𝑗 (𝑧) = ([0, … , 1, … , 0], (0, … , 𝑧, … , 0)),

334 10. The Kodaira Embedding Theorem

with the nonzero entries in the 𝑗th positions. Then 𝜆𝑗 (𝑧) ∈ 𝑆 if and only if 𝑧 = 0,
and
(𝜋1∗ 𝑠0 )(𝜆𝑗 (𝑧)) = ([0, … , 1, … , 0], 𝑧𝑓𝑗 (0, … , 𝑧, … , 0)𝑠(0, … , 𝑧, … , 0)𝑘 ).

The assumption that 𝜋1∗ 𝑠0 ∈ ℐ𝑆2 (𝑊 ̃; 𝐿̃ 𝑘 ) implies that 𝑓𝑗 (0) = 0. Since this is true
for each 𝑗, it follows that 𝑠0 ∈ ℐ{𝑝}
2
(𝑊0 ; 𝐿𝑘 ), and thus 𝜎(𝑝) = 0.
The map Π ̃ 𝑀 in (10.19) is surjective provided 𝐻 1 (𝑀𝑝 ; ℐ 2 (𝐿
̃ 𝑘 )) = 0. Propo-
𝑆
̃ 𝑘 ) ≅ 𝒪(𝐿
̃ 𝑘 ⊗ 𝐿∗ ⊗ 𝐿∗ ), and Corollary 10.8 shows that
𝑝
sition 5.16 shows that ℐ𝑆2 (𝐿 𝑆 𝑆
𝐻 1 (𝑀𝑝 ; 𝒪(𝐿̃ 𝑘 ⊗ 𝐿∗ ⊗ 𝐿∗ )) = 0 provided 𝐾 ̃∗ ⊗ 𝐿̃ 𝑘 ⊗ 𝐿∗ ⊗ 𝐿∗ is positive. By
𝑆 𝑆 𝑆 𝑆
Proposition 10.11, this bundle is isomorphic to
𝜋 ∗ (𝐾 ∗ ⊗ 𝐿𝑘 ) ⊗ (𝐿∗𝑆 )𝑛+1 .
Because 𝑘 ≥ 2𝑛𝑘0 + 𝑘1 ≥ (𝑛 + 1)𝑘0 + 𝑘1 , we can write 𝑘 = (𝑛 + 1)𝑘0 + 𝑘″ with
𝑘″ ≥ 𝑘1 , and rewrite the tensor product above as

𝜋 ∗ (𝐾 ∗ ⊗ 𝐿𝑘 ) ⊗ (𝜋 ∗ 𝐿𝑘0 × 𝐿∗𝑆 )𝑛+1 .

″

̃ 𝑀 is surjective, and it follows as

As above, this is a positive line bundle. Thus Π 𝑝
before that Π𝑀 is also surjective. This completes the proof that 𝒪(𝑀; 𝐿𝑘 ) separates
directions, and therefore 𝐿 is ample. □

emphasis on the Kähler geometry of 𝑀. We say that a Kähler metric on 𝑀 is a

There is another way to express the Kodaira embedding theorem that puts the

Hodge metric if its Kähler class is integral; that is, its image in 𝐻 2 (𝑀; ℝ) lies in
the image of the coefficient homomorphism 𝐻 2 (𝑀; ℤ) → 𝐻 2 (𝑀; ℝ).
Theorem 10.13 (Kodaira Embedding Theorem, Geometric Version). A com-
pact complex manifold is projective if and only if it admits a Hodge metric.

Proof. Suppose first that 𝑀 is projective, so there is a holomorphic embedding

𝑀 ↪ ℂℙ𝑁 for some 𝑁. Let 𝑔FS be the Fubini-Study metric on ℂℙ𝑁 . Recall
from Example 8.14 that the Kähler form of 𝑔FS is equal to 2𝑖 Θ𝐻 , where Θ𝐻 is the
curvature of the Chern connection on the hyperplane bundle 𝐻. Thus the metric
𝑔 ̃ = 𝜋1 𝑔FS has Kähler form 2𝜋𝑖 Θ𝐻 , which is the Chern form of the hyperplane bundle
and therefore is integral by Theorem 7.14. The pullback of this metric to 𝑀 is a
Hodge metric on 𝑀. (This is the reason some authors define the Fubini–Study
metric with the additional factor of 1/𝜋.)
Conversely, suppose 𝑀 admits a Hodge metric 𝑔, and let 𝜔 be its Kähler form.
By the Lefschetz theorem on (1, 1)-classes, there is a line bundle 𝐿 → 𝑀 whose
first real Chern class is represented by 𝜔. Since 𝜔 is a positive (1, 1)-form, this
implies 𝐿 is a positive line bundle, so it follows from Theorem 10.12 that 𝑀 is
projective. □
Applications of the Embedding Theorem 335

Applications of the Embedding Theorem

The Kodaira embedding theorem has numerous applications in complex differential
geometry and algebraic geometry. Here we consider a few of the most important
ones; some others are described in the problems at the end of the chapter.
Theorem 10.14 (Blowups of Projective Manifolds Are Projective). Suppose 𝑀
̃ is the blowup of 𝑀 at finitely many points.
is a projective complex manifold and 𝑀
̃
Then 𝑀 is also projective.

of blowups one point at a time, it suffices to prove this when 𝜋𝑝 ∶ 𝑀𝑝 → 𝑀 is a

Proof. Since a blowup at multiple points can be viewed as the result of a sequence

blowup at one point. Since 𝑀 is projective, it has a positive complex line bundle
𝐿 → 𝑀. Proposition 10.9 shows that for some 𝑘 > 0, the bundle 𝜋𝑝∗ 𝐿𝑘 ⊗ 𝐿∗𝑆 →
𝑀𝑝 is positive (where 𝑆𝑝 ⊆ 𝑀𝑝 is the exceptional hypersurface), so 𝑀𝑝 is also
𝑝

projective.
Theorem 10.15 (Covering Manifolds Are Projective). Suppose 𝑀 ̃ and 𝑀 are
□

compact complex manifolds and 𝜋 ∶ 𝑀 ̃ → 𝑀 is a holomorphic covering map.

̃ is projective if and only if 𝑀 is projective.
Then 𝑀

Proof. First suppose 𝑀 is projective. Then it admits a positive line bundle 𝐿, and
𝜋 ∗ 𝐿 is positive line bundle on 𝑀 ̃ by Proposition 10.5.
Conversely, suppose 𝑀 ̃ is projective, and let 𝑔 ̃ be a Hodge metric on 𝑀 ̃ with
Kähler form 𝜔. ̃
̃ Because 𝑀 and 𝑀 are compact, the covering 𝜋 has only finitely
many sheets. Define a positive (1, 1)-form on 𝑀 as follows. Given 𝑝 ∈ 𝑀, let 𝑈 be
a connected evenly covered neighborhood of 𝑝, and let 𝑠1 , … 𝑠𝑚 ∶ 𝑈 → 𝜋 −1 (𝑈 ) be
the distinct local sections of 𝜋, where 𝑚 is the number of sheets of 𝜋. Define 𝜔|𝑈 =
∑𝑚 𝑖=1 𝑠𝑖 𝜔.
̃ Given any other such neighborhood 𝑈 ′ with sections 𝑠′1 , … , 𝑠′𝑚 , let 𝑈0
∗

be the connected component of 𝑈 ∩ 𝑈 ′ containing 𝑝. Because two local sections on

a connected open set that agree at a point must agree everywhere [LeeTM, Thm.

set after reordering, so 𝜔 is well defined. It satisfies 𝜋 ∗ 𝜔 = 𝑚𝜔.̃

11.12], the restrictions of the first set of sections must agree with those of the second

To see that 𝜔 is integral, by Lemma 6.27 we just need to show that its integral
over every smooth 2-cycle is an integer. Let 𝑐 ∈ Sing∞ 2 (𝑀) be such a cycle. By
[LeeTM, Prop. 13.19], we can replace 𝑐 by a homologous smooth singular cycle
(still denoted by 𝑐) with the property that the image of every singular simplex in 𝑐 is
contained in an evenly covered open set. Writing 𝑐 = ∑𝑗 𝑛𝑗 𝜎𝑗 , we define a smooth
singular chain 𝜋 # 𝑐 ∈ Sing2∞ (𝑀 ̃ ) by

𝜋#𝑐 = 𝑛𝑗 (𝑠1 ∘ 𝜎𝑗 + ⋯ + 𝑠𝑚 ∘ 𝜎𝑗 ),
∑
𝑗

where for each 𝜎𝑗 , we let 𝑠1 , … , 𝑠𝑚 denote the local sections on a neighborhood of

the image of 𝜎𝑗 in some order. Because 𝜋 # (𝜕𝑐) = 𝜕(𝜋 # 𝑐), it follows that 𝜋 # 𝑐 is a
336 10. The Kodaira Embedding Theorem

̃ We compute
cycle in 𝑀.

𝜔= 𝑛𝑗 𝜔= 𝑛𝑗 ( 𝑠∗1 𝜔
̃ + ⋯ + 𝑠∗𝑚 𝜔
̃)
∫𝑐
∑ ∫𝜎𝑗
∑ ∫𝜎𝑗
𝑗 𝑗

= 𝑛𝑗 ( 𝜔
̃+⋯+ 𝜔
̃)
∑ ∫𝑠1 ∘𝜎𝑗 ∫𝑠𝑚 ∘𝜎𝑗
𝑗

= 𝜔.
̃
∫𝜋#𝑐
This is an integer by our assumption on 𝜔.
̃ □
Theorem 10.16. If 𝑀 is a compact Kähler manifold with ℎ2,0 (𝑀) = 0, then 𝑀 is
projective.

Proof. Suppose 𝑀 satisfies the hypothesis. By Hodge symmetry, ℎ0,2 (𝑀) is also
zero, and thus 𝐻dR 2
(𝑀; ℂ) ≅ 𝐻 1,1 (𝑀) by the Hodge decomposition theorem.
Let 𝜔 be a Kähler form on 𝑀. Lemma 6.27 shows that the integral classes span
𝐻dR2
(𝑀; ℝ) over ℝ, so rational linear combinations of integral classes are dense in
𝐻dR (𝑀; ℝ). Therefore by perturbing 𝜔 slightly, we can find a closed real (1, 1)-
2

form 𝜔 ̃ that is a rational linear combination of integral forms and is still positive.
Then multiplying 𝜔 ̃ by a suitable positive integer, we obtain a positive integral
(1, 1)-form, which thus determines a Hodge metric on 𝑀. □

jective. Problem 10-3 shows that every 1-dimensional complex torus is projective,
For our final application, we address the question of which complex tori are pro-

Suppose 𝑉 is an 𝑛-dimensional complex vector space and Λ ⊆ 𝑉 is a lattice.

but in higher dimensions that is no longer the case.

Let 𝐽 ∶ 𝑉 → 𝑉 be the associated complex structure map. A Riemann form for 𝚲

is an antisymmetric real bilinear form Ω on 𝑉 that satisfies
(i) Ω(𝐽 𝑣, 𝐽 𝑤) = Ω(𝑣, 𝑤) for all 𝑣, 𝑤 ∈ 𝑉 ;
(ii) Ω(𝑣, 𝐽 𝑣) > 0 for all nonzero 𝑣 ∈ 𝑉 ;
(iii) Ω(𝑣, 𝑤) ∈ ℤ for all 𝑣, 𝑤 ∈ Λ.
Theorem 10.17 (Characterization of Projective Tori). Suppose 𝑉 is an 𝑛-dimen-
sional complex vector space and Λ ⊆ 𝑉 is a lattice. The torus 𝑀 = 𝑉 /Λ is
projective if and only if there is a Riemann form for Λ.

Proof. Suppose first that Ω is a Riemann form for Λ. Under the canonical identi-
fication of each tangent space of 𝑉 with the underlying real vector space of 𝑉 , we
may consider Ω to be a constant-coefficient 2-form on 𝑉 . Property (i) implies it is
of type (1, 1) by the result of Problem 4-1, and (ii) shows it is positive. Because it

Λ, so it descends to a Kähler form on 𝑀, which we denote by 𝜔.

has constant coefficients, it is closed; and it is also invariant under translations by
Problems 337

We need to show that 𝜔 defines an integral cohomology class; by Lemma 6.27,

it suffices to show that ∫𝑐 𝜔 ∈ ℤ for every smooth 2-cycle 𝑐 in 𝑀. Let (𝑣1 , … , 𝑣2𝑛 )
be a basis for the free abelian group Λ, which is also a basis for 𝑉 over ℝ. Let
(𝑥1 , … , 𝑥2𝑛 ) denote the corresponding real coordinates on 𝑉 , so 𝑣𝑖 has coordinates
(0, … , 0, 1, 0, … , 0) with 1 in the 𝑖th place. For each pair of indices 𝑖 < 𝑗, the
real subspace of 𝑉 spanned by (𝑣𝑖 , 𝑣𝑗 ) projects to an embedded 2-torus 𝑇𝑖𝑗 ⊆ 𝑀. It

that 𝐻2 (𝑀) is a free abelian group of rank (2𝑛2)

generated by the images of 𝐻2 (𝑇𝑖𝑗 )
follows from the Künneth theorem for homology [Hat02, Thm. 3B.6] and induction

torus. Thus we need to show that ∫𝑇 𝜔 ∈ ℤ for each 𝑖 < 𝑗. We can parametrize 𝑇𝑖𝑗
under inclusion, which we can represent as smooth cycles by triangulating each such
𝑖𝑗
by the map 𝜑𝑖𝑗 ∶ [0, 1]2 → 𝑀 defined by 𝜑𝑖𝑗 = 𝜋 ∘ Φ𝑖𝑗 , where 𝜋 ∶ 𝑉 → 𝑀 is the
quotient map and Φ𝑖𝑗 ∶ [0, 1]2 → 𝑉 is given by Φ𝑖𝑗 (𝑠, 𝑡) = (0, … , 𝑠, … , 𝑡, … , 0),
with 𝑠 and 𝑡 in positions 𝑖 and 𝑗. Writing Ω = ∑𝑘𝑙 Ω𝑘𝑙 𝑑𝑥𝑘 ∧𝑑𝑥𝑙 in these coordinates,
we see that 𝜑∗𝑖𝑗 𝜔 = Φ∗𝑖𝑗 Ω = Ω𝑖𝑗 𝑑𝑠 ∧ 𝑑𝑡, and therefore

𝜔= 𝜑∗𝑖𝑗 𝜔 = Ω𝑖𝑗 𝑑𝑠 ∧ 𝑑𝑡 = Ω𝑖𝑗 = Ω(𝑣𝑖 , 𝑣𝑗 ),

∫𝑇𝑖𝑗 ∫[0,1]2 ∫[0,1]2

which is an integer. Thus 𝜔 defines an integral cohomology class, showing that 𝑀

Conversely, suppose 𝑀 is projective. Let 𝑔 be a Hodge metric on 𝑀, and let 𝜔

is projective by the Kodaira embedding theorem.

be its Kähler form. Since Λ is a discrete subgroup of the Lie group 𝑉 , the quotient
space 𝑀 = 𝑉 /Λ is a complex Lie group, which acts transitively on itself by left
multiplication. Define a new 2-form Ω on 𝑀 by

Ω𝑝 (𝑋, 𝑌 ) = (𝛾 ∗ 𝜔)𝑝 (𝑋, 𝑌 )𝑑𝑉 (𝛾),

∫𝑀
where 𝑑𝑉 is a left-invariant volume form on 𝑀. Then it is straightforward to
show that Ω is invariant under left multiplication on 𝑀 and thus lifts to a constant-
coefficient 2-form on 𝑉 , which is a Riemann form for Λ. □

Problem 10-5 gives an example of how to apply this theorem to distinguish pro-
jective and nonprojective tori, and Problem 10-12 applies it to prove that Jacobian
varieties of compact Riemann surfaces are always projective. A projective complex
torus is called an abelian variety. These projective varieties are among the most
intensely studied objects in algebraic geometry.

Problems
10-1. Suppose 𝑀 is a compact complex 𝑛-manifold and 𝐿 → 𝑀 is a negative
line bundle. Prove that 𝐻 𝑞 (𝑀; Ω𝑝 (𝐿)) = 0 for 𝑝 + 𝑞 < 𝑛.
10-2. Prove that a negative line bundle on a compact complex manifold has no
nontrivial holomorphic sections.
338 10. The Kodaira Embedding Theorem

10-3. Show that if 𝑀 is a connected compact Riemann surface of genus 𝑔 and

𝐿 → 𝑀 is a holomorphic line bundle whose degree is greater than 2𝑔,
then 𝐿 is very ample. Use this to prove that every connected compact
Riemann surface is projective. [Hint: Assuming 𝑝, 𝑞 ∈ 𝑀 are distinct
points such that every holomorphic section of 𝐿 vanishing at 𝑝 also van-
ishes at 𝑞, show that 𝒪(𝑀; 𝐿 ⊗ 𝐿∗{𝑝} ) ≅ 𝒪(𝑀; 𝐿 ⊗ 𝐿∗{𝑝} ⊗ 𝐿∗{𝑞} ), and use
Riemann–Roch to derive a contradiction.]

Riemann surface of genus 1 is biholomorphic to a nonsingular cubic curve

10-4. Use the Riemann–Roch theorem to prove that every connected compact

in ℂℙ2 .
10-5. Let 𝑎 ∈ ℝ and let Λ ⊆ ℂ2 be the lattice spanned by the following four
vectors:

𝑣1 = (1, 0), 𝑣2 = (𝑖, 0), 𝑣3 = (0, 1), 𝑣4 = (𝑎, 𝑖).

Show that that the torus ℂ2 /Λ is projective if and only if 𝑎 is rational.

10-6. If 𝐹 ∶ 𝑀 → 𝑁 is a smooth map between compact, connected, oriented,
smooth 𝑛-manifolds, the degree of 𝑭 is the unique integer 𝑘 such that
∫𝑀 𝐹 ∗ 𝜔 = 𝑘 ∫𝑁 𝜔 for every 𝜔 ∈ ℰ 𝑛 (𝑁), and the preimage of every
regular value of 𝐹 contains exactly 𝑘 points counted with signs (+ if 𝐹
is locally orientation-preserving near the point, − if not) [LeeSM, Thm.
17.35]. Suppose 𝑀 is a connected compact Riemann surface of positive

(a) Show that the degree of a holomorphic map 𝐹 ∶ 𝑀 → ℂℙ1 is equal

genus.

to the degree of the line bundle 𝐹 ∗ 𝐻 → 𝑀, where 𝐻 → ℂℙ1 is the

(b) Show that there is a holomorphic map 𝐹 ∶ 𝑀 → ℂℙ1 of degree 2 if

hyperplane bundle.

and only if there exists a holomorphic line bundle 𝐿 → 𝑀 of degree

2 with dim 𝒪(𝑀; 𝐿) = 2. [Hint: Given such a bundle, you will have
to show that it has no base points. Assuming 𝑝 is a base point of 𝐿,
show that 𝒪(𝑀; 𝐿) ≅ 𝒪(𝑀; 𝐿 ⊗ 𝐿∗{𝑝} ) and use Proposition 7.25.]
(c) Show that when 𝑀 has genus 1, it always admits such a map of degree
2.
10-7. A connected compact Riemann surface 𝑀 of genus 𝑔 > 1 is called a
hyperelliptic curve if there is a holomorphic map 𝑀 → ℂℙ1 of degree 2,

bundle 𝐿 → 𝑀 of degree 2 with dim 𝒪(𝑀; 𝐿) = 2. (The name reflects

or equivalently by the result of Problem 10-6, if there is a holomorphic line

Riemann surfaces of genus 1, as shown in Problem 10-6.)

the fact that this generalizes a property shared by all elliptic curves, i.e.,

(a) Show that if 𝑀 has genus 2, then it is hyperelliptic.

Problems 339

(b) Because the canonical bundle 𝐾 → 𝑀 has no base points (Problem

9-15) and dim 𝒪(𝑀; 𝐾) = 𝑔, there is a holomorphic map 𝑀 →
ℂℙ𝑔−1 associated with 𝐾, called the canonical map. Any variety
in ℂℙ𝑔−1 that is the image of a canonical map is called a canonical

and only if 𝑀 is not hyperelliptic.

curve. Show that the canonical map is a holomorphic embedding if

(c) Show that every nonsingular quartic curve in ℂℙ2 is a canonical

curve, and thus not hyperelliptic.
10-8. Suppose (𝑀, 𝑔) is a compact Kähler manifold whose Ricci curvature is
either positive-definite or negative-definite. Prove that 𝑀 is projective.
10-9. Let 𝐻 → ℂℙ𝑛 be the hyperplane bundle. Prove the following:

⎧ 𝑛+𝑑 , 𝑞 = 0 and 𝑑 ≥ 0,
⎪( 𝑛 )
⎪
dim 𝐻 𝑞 (ℂℙ𝑛 ; 𝒪(𝐻 𝑑 )) = ⎨ −𝑑 − 1 , 𝑞 = 𝑛 and 𝑑 ≤ −𝑛 − 1,
⎪( 𝑛 )
⎪0
⎩ otherwise.
10-10. Suppose (𝑀, 𝑔) is a connected compact 𝑛-dimensional Kähler manifold.
For any 𝑝 ∈ 𝑀, let Hol(𝑝) denote the holonomy group at 𝑝 (see Problem
8-19). Show that 𝑀 is a Calabi–Yau manifold and 𝑔 is Ricci-flat if and
only if Hol(𝑝) ⊆ SU(𝑛) for some 𝑝 ∈ 𝑀 and an appropriate choice of
basis for 𝑇𝑝 𝑀.
10-11. Show that if (𝑀, 𝑔) is a connected Ricci-flat Calabi–Yau manifold of di-
mension 𝑛 ≥ 3 whose holonomy group is equal to all of SU(𝑛) for some
𝑝 ∈ 𝑀, then ℎ2,0 (𝑀) = 0 and therefore 𝑀 is projective. [Hint: First use
Problem 9-8 to show that if 𝜂 is a harmonic (2, 0)-form, then 𝜂𝑝 is invari-
ant under the holonomy group Hol(𝑝). Then show that Hol(𝑝) ≅ SU(𝑛)
acts transitively on the set of 2-dimensional complex-linear subspaces of
𝑇𝑝′ 𝑀, but there is a 2-dimensional subspace 𝑉 such that 𝜂𝑝 |𝑉 ×𝑉 ≡ 0.]
[Remark: Because of this result, some authors define Calabi–Yau mani-
folds as Kähler manifolds with holonomy equal to SU(𝑛).]

pose 𝑀 is a connected compact Riemann surface of genus 𝑔 ≥ 1, and

10-12. This problem outlines a proof that Jacobian varieties are projective. Sup-

Jac(𝑀) = Ω1 (𝑀)∗ /Λ is its Jacobian, where Λ is the lattice 𝜑(𝐻1 (𝑀))

defined by (9.50). Let (𝑎1 , … , 𝑎𝑔 , 𝑏1 , … , 𝑏𝑔 ) be the cycles represent-
ing a basis for 𝐻1 (𝑀) defined in the proof of Theorem 9.68, and let
(𝛼 1 , … , 𝛼 𝑔 , 𝛽 1 , … , 𝛽 𝑔 ) be closed 1-forms satisfying (9.51)–(9.54).
(a) For any closed 1-forms 𝜃 and 𝜃 ̃ on 𝑀, show that
𝑔
𝜃 ∧ 𝜃̃ = 𝜃 )( 𝜃 )̃ − ̃
(∫ 𝜃 )(∫ 𝜃 ).
∫𝑀
∑(∫𝑎𝑗 ∫𝑏𝑗 𝑏𝑗 𝑎𝑗
𝑗=1
340 10. The Kodaira Embedding Theorem

[Hint: First show both sides of the equation are unchanged if 𝜃 is

replaced by a cohomologous form. Then write 𝜃 (mod exact forms)
as a linear combination of the 𝛼 𝑗 ’s and 𝛽 𝑗 ’s and use (9.51)–(9.54).]
(b) Choose a basis (𝜂 1 , … , 𝜂 𝑔 ) for Ω1 (𝑀) over ℂ, and let (𝜀1 , … , 𝜀𝑔 ) be
the dual basis. Write the period matrix of 𝑀 as Π = (𝐴 𝐵), where

𝐴𝑘𝑗 = 𝜂𝑘, 𝐵𝑗𝑘 = 𝜂𝑘.

∫𝑎𝑗 ∫𝑏𝑗

By applying part (a) to 𝜂 𝑘 ∧ 𝜂 𝑙 and 𝜂 𝑘 ∧ ∗𝜂 𝑙 , prove that the periods

(i) For all 𝑘, 𝑙 = 1, … , 𝑔,

satisfy the following Riemann bilinear relations:

𝑔
𝐴𝑘𝑗 𝐵𝑗𝑙 − 𝐵𝑗𝑘 𝐴𝑙𝑗 ) = 0.
∑(
𝑗=1

(ii) The following Hermitian matrix is positive definite:

𝑔
𝑄 =𝑖 𝐴𝑘 𝐵 𝑙 − 𝐵𝑗𝑘 𝐴𝑙𝑗 ).
∑( 𝑗 𝑗
𝑘𝑙

𝑗=1

(In matrix notation, these can be written succinctly as 𝐴𝐵 𝑇 − 𝐵𝐴𝑇 =

0 and 𝑄 = 𝑖(𝐴𝐵 𝑇 − 𝐵𝐴𝑇 ) >> 0.)
(c) Show that the matrix 𝐴 is nonsingular, and conclude that by choos-
ing a new basis 𝜂 𝑚
̃ = ∑𝑘 (𝐴−1 )𝑚𝑘 𝜂 for Ω (𝑀), we can put the period
𝑘 1

matrix into the form (𝐼 𝑍) for some 𝑔 × 𝑔 matrix 𝑍 that is sym-

metric and has positive definite imaginary part. [Hint: Assuming 𝐴
is singular, show that there is a nonzero column matrix 𝑤 such that
𝑤𝑇 𝑄𝑤 = 0.]
(d) Define a real-linear isomorphism Φ ∶ ℝ2𝑔 → ℂ𝑔 by
Φ(𝑥1 , … , 𝑥𝑔 , 𝑦1 , … , 𝑦𝑔 ) = (𝑧1 , … , 𝑧𝑔 ), where
𝑔 𝑔
𝑧 =
𝛼
𝑥
𝑗
𝐴𝛼𝑗 + 𝑦𝑗 𝐵𝑗𝛼 ,
∑ ∑
𝑗=1 𝑗=1

and let Ψ = Φ−1 ∶ ℂ𝑔 → ℝ2𝑔 . With (𝐴 𝐵) = (𝐼 𝑍) as above, show

that the differential of Ψ satisfies
𝑔
𝜕 𝑗 𝜕 𝑗 𝜕
𝐷Ψ ( = 𝐶 + 𝐷𝛼 𝑗 ,
𝜕𝑧𝛼 ) 𝛼=1
∑ 𝛼 𝜕𝑥𝑗 𝜕𝑦
𝑔
𝜕 𝑗 𝜕 𝑗 𝜕
𝐷Ψ ( 𝛼 ) = 𝐶𝛼 𝑗 + 𝐷𝛼 𝑗 ,
𝜕𝑧 ∑ 𝜕𝑥 𝜕𝑦
𝛼=1
Problems 341

where the 𝑔 × 𝑔 matrices 𝐶 and 𝐷 are given by 𝐶 = −𝑍(𝑍 − 𝑍)−1 ,

𝐷 = (𝑍 − 𝑍)−1 . Use these formulas to show that 𝐶 𝑇 𝐷 − 𝐷𝑇 𝐶 = 0
and −𝑖(𝐶 𝑇 𝐷 − 𝐷𝑇 𝐶) is positive definite.
(e) Let Ω0 be the 2-form ∑𝑗=1 𝑑𝑥𝑗 ∧ 𝑑𝑦𝑗 on ℝ2𝑔 , and let Ω = Ψ∗ Ω0 .
𝑔

Show that Ω is a Riemann form for Λ, and therefore Jac(𝑀) is pro-

jective.
Bibliography

[AN54] Yasuo Akizuki and Shigeo Nakano, Note on Kodaira-Spencer’s proof of Lefschetz theorems, Proc. Japan
Acad. 30 (1954), 266–272. MR0066694
[Aub78] Thierry Aubin, Équations du type Monge-Ampère sur les variétés kählériennes compactes (French,
with English summary), Bull. Sci. Math. (2) 102 (1978), no. 1, 63–95. MR0494932
[Bal06] Werner Ballmann, Lectures on Kähler Manifolds, ESI Lectures in Mathematics and Physics, European
Mathematical Society (EMS), Zürich, 2006, DOI 10.4171/025. MR2243012
[Bes87] Arthur L. Besse, Einstein Manifolds, Ergebnisse der Mathematik und ihrer Grenzgebiete (3) [Results
in Mathematics and Related Areas (3)], vol. 10, Springer, New York, 1987, DOI 10.1007/978-3-540-
74311-8. MR0867684
[Boc46] Salomon Bochner, Vector fields and Ricci curvature, Bull. Amer. Math. Soc. 52 (1946), 776–797, DOI
10.1090/S0002-9904-1946-08647-4. MR0018022
[BS53] Armand Borel and Jean-Pierre Serre, Groupes de Lie et puissances réduites de Steenrod (French), Amer.
J. Math. 75 (1953), 409–448, DOI 10.2307/2372495. MR0058213
[Bre97] Glen E. Bredon, Sheaf Theory, 2nd ed., Graduate Texts in Mathematics, vol. 170, Springer, New York,
1997, DOI 10.1007/978-1-4612-0647-7. MR1481706
[BC13] James Ward Brown and Ruel V. Churchill, Complex Variables and Applications, 9th ed., McGraw–Hill
Book Co., New York, 2013.
[Cal57] Eugenio Calabi, On Kähler manifolds with vanishing canonical class, Algebraic geometry and topol-
ogy. A symposium in honor of S. Lefschetz, Princeton Univ. Press, Princeton, NJ, 1957, pp. 78–89.
MR0085583
[CdS01] Ana Cannas da Silva, Lectures on Symplectic Geometry, Lecture Notes in Mathematics, vol. 1764,
Springer, New York, 2001, DOI 10.1007/978-3-540-45330-7. MR1853077
[Car53] Henri Cartan, Variétés analytiques complexes et cohomologie (French), Colloque sur les Fonctions de
Plusieurs Variables, Tenu à Bruxelles, 1953, Georges Thone, Liège, 1953, pp. 41–55. MR0064154
[Čec32] Eduard Čech, Théorie générale de l’homologie dans un espace quelconque (French), Fund. Math. 19
(1932), 149–183, DOI 10.4064/fm-19-1-184-197.
[Che46] Shiing-Shen Chern, Characteristic classes of Hermitian manifolds, Ann. of Math. (2) 47 (1946), 85–
121, DOI 10.2307/1969037. MR0015793
[Cho49] Wei-Liang Chow, On compact complex analytic varieties, Amer. J. Math. 71 (1949), 893–914, DOI
10.2307/2372375. MR0033093

343
344 Bibliography

[Dem12] Jean-Pierre Demailly, Complex Analytic and Differential Geometry (2012), available at https://
[Link]/~demailly/[Link].
[Dol53] Pierre Dolbeault, Sur la cohomologie des variétés analytiques complexes (French), C. R. Acad. Sci.
Paris 236 (1953), 175–177. MR0052771
[Dol56] , Formes différentielles et cohomologie sur une variété analytique complexe. I (French), Ann.
of Math. (2) 64 (1956), 83–130, DOI 10.2307/1969950. MR0083166
[FK92] Hershel M. Farkas and Irwin Kra, Riemann Surfaces, 2nd ed., Graduate Texts in Mathematics, vol. 71,
Springer-Verlag, New York, 1992, DOI 10.1007/978-1-4612-2034-3. MR1139765
[Fub04] Guido Fubini, Sulle metriche definite da una forma Hermitiana (Italian), Atti del Reale Istituto Veneto
di Scienze, Lettere ed Arti 63 (1904), 501–513.
[Gam01] Theodore W. Gamelin, Complex Analysis, Undergraduate Texts in Mathematics, Springer, New York,
2001, DOI 10.1007/978-0-387-21607-2. MR1830078
[GKZ08] Israel M. Gelfand, Mikhail M. Kapranov, and Andrei V. Zelevinsky, Discriminants, Resultants and
Multidimensional Determinants, Modern Birkhäuser Classics, Birkhäuser Boston, Inc., Boston, MA,
2008. Reprint of the 1994 edition. MR2394437
[God73] Roger Godement, Topologie Algébrique et Théorie des Faisceaux (French), Publications de l’Institut
de Mathématique de l’Université de Strasbourg, vol. XIII, Hermann, Paris, 1973. Troisième édition
revue et corrigée; Actualités Scientifiques et Industrielles, No. 1252. [Current Scientific and Industrial
Topics]. MR0345092
[GH94] Phillip Griffiths and Joseph Harris, Principles of Algebraic Geometry, Wiley Classics Library, John
Wiley & Sons, Inc., New York, 1994. Reprint of the 1978 original, DOI 10.1002/9781118032527.
MR1288523
[Gro57] Alexander Grothendieck, Sur quelques points d’algèbre homologique (French), Tohoku Math. J. (2) 9
(1957), 119–221, DOI 10.2748/tmj/1178244839. MR0102537
[GR09] Robert C. Gunning and Hugo Rossi, Analytic Functions of Several Complex Variables, AMS Chelsea
Publishing, Providence, RI, 2009. Reprint of the 1965 original.
[Har06] Friedrich Hartogs, Einige Folgerungen aus der Cauchyschen Integralformel bei Funktionen mehrerer
Veränderlichen (German), Sitzungsberichte der Königlich Bayerischen Akademie der Wissenschaften
zu München, Mathematisch-Physikalische Klasse 36 (1906), 223–242.
[Har77] Robin Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics, No. 52, Springer, New York,
1977. MR0463157
[Hat02] Allen Hatcher, Algebraic Topology, Cambridge University Press, Cambridge, UK, 2002. MR1867354
[Hod41] William V. D. Hodge, The Theory and Applications of Harmonic Integrals, Cambridge University Press,
Cambridge; The Macmillan Company, New York, 1941. MR0003947
[Hör90] Lars Hörmander, An Introduction to Complex Analysis in Several Variables, 3rd ed., North–Holland
Mathematical Library, vol. 7, North–Holland Publishing Co., Amsterdam, 1990. MR1045639
[Huy05] Daniel Huybrechts, Complex Geometry, Universitext, Springer-Verlag, Berlin, 2005. MR2093043
[Jos06] Jürgen Jost, Compact Riemann Surfaces, 3rd ed., Universitext, Springer-Verlag, Berlin, 2006. An in-
troduction to contemporary mathematics, DOI 10.1007/978-3-540-33067-7. MR2247485
[Kod53] Kunihiko Kodaira, On a differential-geometric method in the theory of analytic stacks, Proc. Nat. Acad.
Sci. U.S.A. 39 (1953), 1268–1273, DOI 10.1073/pnas.39.12.1268. MR0066693
[Kod54] , On Kähler varieties of restricted type (an intrinsic characterization of algebraic varieties),
Ann. of Math. (2) 60 (1954), 28–48, DOI 10.2307/1969701. MR0068871
[Kod64] , On the structure of compact complex analytic surfaces. I, Amer. J. Math. 86 (1964), 751–798,
DOI 10.2307/2373157. MR0187255
[Kra01] Steven G. Krantz, Function Theory of Several Complex Variables, AMS Chelsea Publishing, Provi-
dence, RI, 2001. Reprint of the 1992 edition, DOI 10.1090/chel/340. MR1846625
[Lan02] Serge Lang, Algebra, 3rd ed., Graduate Texts in Mathematics, vol. 211, Springer-Verlag, New York,
2002, DOI 10.1007/978-1-4613-0041-0. MR1878556
Bibliography 345

[LeeRM] John M. Lee, Introduction to Riemannian Manifolds, 2nd ed., Graduate Texts in Mathematics, vol. 176,
Springer, Cham, 2018. MR3887684
[LeeSM] , Introduction to Smooth Manifolds, 2nd ed., Graduate Texts in Mathematics, vol. 218, Springer,
New York, 2013. MR2954043
[LeeTM] , Introduction to Topological Manifolds, 2nd ed., Graduate Texts in Mathematics, vol. 202,
Springer, New York, 2011, DOI 10.1007/978-1-4419-7940-7. MR2766102
[Lef24] Solomon Lefschetz, L’analysis Situs et la Géométrie Algébrique, Gauthier-Villars et Cie., Paris, 1924.
[MH98] Jerrold E. Marsden and Michael J. Hoffman, Basic Complex Analysis, 3rd ed., W. H. Freeman and
Company, New York, 1998.
[May99] J. Peter May, A Concise Course in Algebraic Topology, Chicago Lectures in Mathematics, University
of Chicago Press, Chicago, IL, 1999. MR1702278
[Mil00] Haynes Miller, Leray in Oflag XVIIA: the origins of sheaf theory, sheaf cohomology, and spectral
sequences, Gaz. Math. 84 (2000), 17–34. Jean Leray (1906–1998). MR1775587
[Mor07] Andrei Moroianu, Lectures on Kähler Geometry, London Mathematical Society Student Texts, vol. 69,
Cambridge University Press, Cambridge, 2007, DOI 10.1017/CBO9780511618666. MR2325093
[Mun66] James R. Munkres, Elementary Differential Topology, Revised edition, Annals of Mathematics Studies,
vol. No. 54, Princeton University Press, Princeton, NJ, 1966. Lectures given at Massachusetts Institute
of Technology, Fall, 1961. MR0198479
[Mun84] , Elements of Algebraic Topology, Addison-Wesley Publishing Company, Menlo Park, CA,
1984. MR0755006
[Nir73] Louis Nirenberg, Lectures on Linear Partial Differential Equations, Conference Board of the Math-
ematical Sciences Regional Conference Series in Mathematics, vol. No. 17, American Mathematical
Society, Providence, RI, 1973. Expository Lectures from the CBMS Regional Conference held at the
Texas Technological University, Lubbock, Tex., May 22–26, 1972. MR0450755
[Pet22] Dan Petersen, A remark on singular cohomology and sheaf cohomology, Math. Scand. 128 (2022),
no. 2, 229–238. MR4472714
[Rie57] Bernhard Riemann, Theorie der Abel’schen Functionen (German), J. Reine Angew. Math. 54 (1857),
115–155, DOI 10.1515/crll.1857.54.115. MR1579035
[Roc65] Gustav Roch, Ueber die Anzahl der willkürlichen Constanten in algebraischen Functionen (German),
J. Reine Angew. Math. 64 (1865), 372–376, DOI 10.1515/crll.1865.64.372. MR1579304
[Rot08] Gian-Carlo Rota, Indiscrete Thoughts, Modern Birkhäuser Classics, Birkhäuser Boston, Inc., Boston,
MA, 2008. Reprint of the 1997 edition [Birkhäuser Boston, Boston, MA; MR1419503]; With forewords
by Reuben Hersh and Robert Sokolowski; Edited, with notes and an epilogue by Fabrizio Palombi, DOI
10.1007/978-0-8176-4781-0. MR2374113
[Rud76] Walter Rudin, Principles of Mathematical Analysis, 3rd ed., International Series in Pure and Applied
Mathematics, McGraw-Hill Book Co., New York-Auckland-Düsseldorf, 1976. MR0385023
[Sel16] Yehonatan Sella, Comparison of sheaf cohomology and singular cohomology (2016), available at
arXiv:1602.06674 [[Link]].
[Ser55a] Jean-Pierre Serre, Un théorème de dualité (French), Comment. Math. Helv. 29 (1955), 9–26, DOI
10.1007/BF02564268. MR0067489
[Ser55b] , Géométrie algébrique et géométrie analytique (French), Ann. Inst. Fourier (Grenoble) 6
(1955/56), 1–42. MR0082175
[Siu83] Yum-Tong Siu, Every K3 surface is Kähler, Invent. Math. 73 (1983), no. 1, 139–150, DOI
10.1007/BF01393829. MR0707352
[Stu05] Eduard Study, Kürzeste Wege im komplexen Gebiet (German), Math. Ann. 60 (1905), no. 3, 321–378,
DOI 10.1007/BF01457616. MR1511309
[vdW50] Bartel Leendert van der Waerden, Modern Algebra, Vol. II, Frederick Ungar Publishing Co., New York,
NY, 1950. Translated from the second revised German edition by Theodore J. Benac.
346 Bibliography

[War83] Frank W. Warner, Foundations of Differentiable Manifolds and Lie Groups, Graduate Texts in Math-
ematics, vol. 94, Springer-Verlag, New York-Berlin, 1983. Corrected reprint of the 1971 edition.
MR0722297
[Wei52] André Weil, Sur les théorèmes de de Rham (French), Comment. Math. Helv. 26 (1952), 119–145, DOI
10.1007/BF02564296. MR0050280
[Wei14] , Oeuvres Scientifiques/Collected Papers. II. 1951–1964 (French), Springer Collected Works in
Mathematics, Springer, Heidelberg, 2014. Reprint of the 2009 [MR2883739] and 1979 [MR0537935]
editions. MR3309921
[Wel08] Raymond O. Wells Jr., Differential Analysis on Complex Manifolds, 3rd ed., Graduate Texts in Math-
ematics, vol. 65, Springer, New York, 2008. With a new appendix by Oscar Garcia-Prada, DOI
10.1007/978-0-387-73892-5. MR2359489
[Yau78] Shing-Tung Yau, On the Ricci curvature of a compact Kähler manifold and the complex Monge-Ampère
equation. I, Comm. Pure Appl. Math. 31 (1978), no. 3, 339–411, DOI 10.1002/cpa.3160310304.
MR0480350
[YN11] Shing-Tung Yau and Steve Nadis, String theory and the geometry of the universe’s hidden dimensions,
Notices Amer. Math. Soc. 58 (2011), no. 8, 1067–1076. MR2856140
[Zhe00] Fangyang Zheng, Complex Differential Geometry, AMS/IP Studies in Advanced Mathematics, vol. 18,
American Mathematical Society, Providence, RI; International Press, Boston, MA, 2000, DOI
10.1090/amsip/018. MR1777835
Notation Index

∗ (Hodge star operator), 263 𝑐1ℝ (𝐸) (first real Chern class), 203
∞ (point at infinity), 56 𝐶𝑔 (conjugation by 𝑔), 100

⟨⋅, ⋅⟩ (Hermitian inner product), 76

♯ (sharp operator), 229 𝑐(𝐿) (sheaf-theoretic Chern class), 181

⟨⋅, ⋅⟩ (Kronecker pairing), 166

𝐶 𝑝 (𝒰; 𝒮 ) (group of Čech cochains), 147

(⋅, ⋅) (global Hodge inner product), 257, 258

‖⋅‖ (global Hodge norm), 258
ch(𝐾) (convex hull), 67

[⋅, ⋅] (commutator bracket), 281

Cl0 (𝑀) (divisor classes of degree 0), 296
Cl(𝑀) (divisor class group), 94

[⋅] (singular homology class), 165

ℂℙ0 (0-dimensional projective space), 5

[⋅] (cohomology class of a Čech cocycle),

ℂℙ𝑛 (complex projective space), 5
(ℂℙ𝑛 )∗ (dual projective space), 54

[[⋅]] (equivalence class in direct limit), 153

149 ℂℙ𝑛 (ℂℙ𝑛 with opposite orientation), 101

𝜕 (conjugate Dolbeault operator), 105

⅃ (interior multiplication), 43
≅ (isomorphic bundles), 25
𝜕 (singular boundary operator), 165
≅ (isomorphic sheaves), 126
𝜕 (Dolbeault operator), 105
𝐴∗ (Hermitian adjoint of a matrix), 68 𝜕𝑓 /𝜕𝑧𝑗 (complex partial derivative), 13
𝐴∗ (adjoint of a linear map), 258 𝜕𝑓 /𝜕𝑧𝑗 (complex vector field), 27
𝐴∗ (cochain complex), 107 𝜕𝑓 /𝜕𝑧𝑗 (complex vector field), 27
Alb(𝑀) (Albanese variety), 300 𝜕𝑓 /𝜕𝑥𝑗 (derivative of complex-valued

♭ (flat operator), 229 𝜕 𝐸 (Cauchy–Riemann operator on

function), 17

𝔹2𝑛 (unit ball), 4

𝑏𝑘 (𝑀) (Betti number), 108 ∇ (connection), 193
bundle-valued forms), 116

𝐵(𝐿) (base locus), 95 ∇𝜎 (total covariant derivative of a section),

𝐵 𝑝 (𝒰; 𝒮 ) (group of Čech coboundaries), 149
𝐵𝑟 (𝑝) (open ball of radius 𝑟), 4 𝛿 (Čech coboundary operator), 148
194

𝐵(𝑍, 𝑊 ) (holomorphic bisectional 𝛿 (singular coboundary operator), 166

Δ (Laplace–Beltrami operator), 265
Δ𝑑 (Hodge Laplacian), 269
curvature, 253

𝒞 (sheaf of continuous functions), 123 Δ𝜕 (Dolbeault Laplacian), 274

𝒞 ∗ (sheaf of nonvanishing continuous Δ𝜕 (conjugate Dolbeault Laplacian), 274
Δ𝑘 (standard simplex), 164
𝑐1 (∇) (Chern form of a connection), 202 𝐷 (exterior covariant derivative), 200
functions), 123

347
348 Notation Index

𝔻 (unit disk), 4 𝐹 ∗ (induced homomorphism on Dolbeault

𝐷𝐹 (differential of a smooth map), 30
𝐷′ 𝐹 (holomorphic Jacobian), 31, 37
cohomology), 108

𝐷𝐹 (𝑝) (differential of a smooth map), 30 Γ(𝐸) (space of smooth sections of 𝐸), 25

𝐷𝑟𝑛 (𝑝) (polydisk of radius 𝑟), 4 Γ𝑐 (𝐸) (space of compactly supported
𝐷𝑟 (𝑝) (disk of radius 𝑟), 4
𝑑 ∗ (formal adjoint of 𝑑), 262 𝐺 (constant sheaf with coefficients in 𝐺), 123
sections), 259

𝐷𝑡 (covariant derivative along a curve), 197 𝐺/Γ (coset space), 10

G𝑘 (𝑉 ) (Grassmannian), 6
𝐺𝑝 (skyscraper sheaf), 123
dimℂ (complex dimension), 3

𝑔CH (complex hyperbolic metric), 243

dimℝ (real dimension), 3

𝑔E (standard metric on ℂ𝑛 ), 225

div (divergence), 265

Div0 (𝑀) (divisors of degree 0), 296, 311 𝑔FS (Fubini–Study metric), 235
Div(𝑀) (group of divisors), 93

𝑑𝑉𝑔 (Riemannian volume form), 237 GL(𝑛, ℂ) (group of invertible complex

GL(𝑉 ) (group of linear automorphisms of

matrices), 10
𝜀 (exponential sheaf morphism), 127
𝑉 ), 10
𝜀𝐽 (wedge of basis 1-forms), 262
ℰ (sheaf of smooth complex-valued
grad (gradient), 265

𝐻 (hyperplane bundle), 88, 90

ℰ ∗ (sheaf of nonvanishing smooth
functions), 123
𝐻 𝑑 (power of the hyperplane bundle), 88
𝐻dR
𝑘
(𝑀) (de Rham cohomology), 108
ℰ (𝐸) (sheaf of smooth sections of 𝐸), 123
functions), 123
𝐻dR (𝑀; ℂ) (complex de Rham
𝑘
𝑒𝑗 (standard basis for ℂ𝑛 ), 74
ℰ 𝑘 (sheaf of smooth complex-valued 𝐻dR (𝑀; ℝ) (real de Rham cohomology), 143
cohomology), 143
𝑘
𝑘-forms), 123 𝐻𝑘∞ (𝑀) (homology of smooth chains), 175
ℰ 𝑘 (𝐸) (sheaf of bundle-valued 𝑘-forms), 123 𝐻𝑘 (𝑀) (singular homology), 165
ℰ 𝑘 (𝑀) (sections of Λ𝑘ℂ 𝑀), 104 𝐻Sing,∞
𝑘
(𝑀; 𝐺) (cohomology of smooth
ℰ 𝑝,𝑞 (sheaf of smooth (𝑝, 𝑞)-forms), 124
ℰ 𝑝,𝑞 (𝐸) (sheaf of smooth 𝐸-valued 𝐻Sing (𝑀; 𝐺) (singular cohomology), 166
cochains), 175
𝑘
(𝑝, 𝑞)-forms), 124 𝐻 (𝑀) (Dolbeault cohomology), 108
𝑝,𝑞
ℰ 𝑝,𝑞 (𝑀) (sections of Λ𝑝,𝑞 𝑀), 104 ℋ 𝑝,𝑞 (𝑀) (harmonic (𝑝, 𝑞)-forms), 275
ℰ 𝑝,𝑞 (𝑀; 𝐸) (bundle-valued (𝑝, 𝑞)-forms), ℎ𝑝,𝑞 (𝑀) (Hodge number), 108
𝐻 𝑝,𝑞 (𝑀; 𝐸) (Dolbeault cohomology with
ℰ 𝑞 (𝑀; 𝐸) (bundle-valued 𝑞-forms), 114
116
coefficients in 𝐸), 117
ℰℝ (sheaf of smooth real-valued functions), ℋ 𝑝,𝑞 (𝑀; 𝐸) (𝜕 𝐸 -harmonic bundle-valued

𝐸 ×𝑀 𝐸 (fiber product), 129

124
𝐻 𝑞 (𝐴∗ ) (cohomology of a cochain complex),
forms), 277

End(𝐸) (endomorphism bundle), 81

ℋ 𝑞 (𝑀) (harmonic 𝑞-forms), 270
107

𝜑𝑓 (section of 𝐻 𝑑 ), 88 𝐻(𝑍) (holomorphic sectional curvature),

(𝑓 ) (divisor of a meromorphic function), 93
𝐹 (morphism of constant sheaves), 127
241

𝐹𝑖,𝑘 (face map), 165 Hom(𝐸, 𝐸 ′ ) (homomorphism bundle), 81

Hol(𝑝) (holonomy group), 254

𝐹𝑝 (stalk homomorphism), 128

[𝑓 ]𝑝 (germ of a function), 127 ℐ𝑆 (ideal sheaf of 𝑆), 124
𝐹# (coefficient homomorphism on singular ℐ𝑆2 (sheaf of holomorphic functions
vanishing to second order on 𝑆), 124
𝐹∗ (coefficient homomorphism in singular ℐ𝑆 (𝐸) (sheaf of holomorphic sections
cochains), 166

vanishing on 𝑆), 124

𝐹∗ (induced sheaf cohomology ℐ𝑆2 (𝐸) (sheaf of holomorphic sections
cohomology), 166

homomorphism), 153 vanishing to second order on 𝑆), 124

Notation Index 349

𝐽 (almost complex structure), 39 𝜌 (Ricci form), 247

𝐽 (complex structure on a vector space), 34 𝑟𝑈𝑉 (restriction map in a presheaf), 121
𝑗 (barred index), 227 𝑅𝑐 (Ricci curvature), 245
𝐽𝑀 (almost complex structure), 36 𝑅𝑚 (Riemann curvature tensor), 239

(𝜎) (divisor of a section of a line bundle), 94,

Jac(𝑀) (Jacobian variety), 305

𝜅 (Kronecker homomorphism), 167

𝐾 (canonical bundle), 118 ∑𝐼 (sum over increasing multi-indices), 105
211
′
𝐾 ∗ (anticanonical bundle), 118 𝑆 (scalar curvature), 245
̂ (holomorphic hull), 65
𝐾 𝒮̂ (sheaf of rough sections), 186
𝐾𝑀 (canonical bundle), 118 [𝑠0 (𝑝), … , 𝑠𝑚 (𝑝)] (associated map), 95
𝐾𝑀∗
𝒮 (𝐾) (sections of 𝒮 over a closed subset,
Ker 𝐹 (kernel of a sheaf morphism), 138
(anticanonical bundle), 118

𝒮 ⊗ℛ 𝒯 (tensor product sheaf), 134

190
Λ𝑘ℂ 𝑀 (bundle of complex 𝑘-forms), 103
Λ𝑝,𝑞 𝑀 (bundle of (𝑝, 𝑞)-forms), 104 𝒮𝑝 (stalk of a sheaf), 128
𝐿ℂ (complexification of a linear map), 23 [𝑠]𝑝 (germ of a section), 128
𝐿𝜔 (Lefschetz operator), 281 𝒮 (𝑈 ) (sections of 𝒮 over an open subset),
𝐿{𝑝} (point bundle), 93
𝒮 |𝑉 (restriction of a sheaf to an open
121
lim 𝐺𝛼 (direct limit), 128
⟶

[𝑀] (fundamental homology class), 165 𝑠|𝑉 (restriction of a section), 122

subset), 123

𝑀/Γ (quotient space by group action), 10 sec(𝑣, 𝑤) (sectional curvature), 252

𝜔CH (complex hyperbolic Kähler form), 243

𝑘 (𝑀) (smooth singular chain group),
sgn (sign of a permutation), 257
Sing∞
𝜔E (standard Kähler form on ℂ𝑛 ), 225
𝜔FS (Fubini–Study Kähler form), 235 Sing𝑘 (𝑀) (singular chain group), 165
175

Ω𝑝 (sheaf of holomorphic 𝑝-forms), 124 Sing𝑘 (𝑀; 𝐺) (singular cochain group), 165
Ω𝑝 (𝐸) (sheaf of holomorphic 𝐸-valued SL(2, ℤ) (integer matrices with determinant
𝑝-forms), 124 1), 42
Ω𝑝 (𝑀) (space of holomorphic 𝑝-forms), 118 SL(𝑛, ℂ) (complex matrices with
Ω𝑝 (𝑀; 𝐸) (space of holomorphic 𝐸-valued determinant 1), 51
𝑝-forms, 118 SO(𝑛, ℂ) (subgroup of O(𝑛, ℂ) with
𝒪 (sheaf of holomorphic functions), 124 determinant 1), 51
𝒪 ∗ (sheaf of nonvanishing holomorphic St 𝑣 (star of a vertex), 191

𝒪(𝑑) (algebraic geometry notation for 𝐻 𝑑 ),

functions), 124
Θ (curvature of a connection), 198
Θ𝑘𝑗 (curvature 2-forms), 199
𝒪(𝐸) (sheaf of nonvanishing sections of 𝐸),
88, 137
𝜃𝑗𝑘 (connection 1-forms), 194
𝑇ℂ 𝑀 (complexified tangent bundle), 26
𝒪(𝑀) (holomorphic functions on 𝑀), 7
124
𝑇ℂ∗ 𝑀 (complexified cotangent bundle), 26
𝒪(𝑀; 𝐸) (space of global holomorphic
𝑇𝐽 𝑀 (𝑇 𝑀 with complex structure), 38
𝑇 ′ 𝑀 (holomorphic tangent bundle), 37
O(𝑛, ℂ) (complex orthogonal matrices), 51
sections), 72
𝑇 ″ 𝑀 (antiholomorphic tangent bundle), 37
𝒪(𝑈 ; 𝐸) (space of local holomorphic
𝑇𝑝′ 𝑀 (holomorphic tangent space), 37
𝑇𝑝″ 𝑀 (antiholomorphic tangent space), 37
sections), 72

𝒫 (sheaf of pluriharmonic functions), 189

𝑃 𝛾 (parallel transport operator), 254
𝑃 ∗ (formal adjoint of 𝑃 ), 259
U(𝑛) (unitary group), 68

ℙ(𝑉 ) (projectivization of 𝑉 ), 6 𝑉 ′ (𝑖-eigenspace of complex structure), 34

𝑉 ″ ((−𝑖)-eigenspace of complex structure),
Pic0 (𝑀) (Picard variety), 182
Pic(𝑀) (Picard group), 83
34
350 Notation Index

𝑉 ∗ (space of conjugate-linear functionals),

𝑉ℂ (complexification of a vector space), 23

311

𝑉ℝ (underlying real vector space), 34

[𝑤] (point in projective space), 5

𝑧𝑗 (holomorphic coordinates), 3
𝑧𝑗 (conjugate of 𝑧𝑗 ), 3
𝑧𝑗 (conjugate of 𝑧𝑗 ), 227
𝒵 𝑘 (sheaf of closed complex-valued
𝑘-forms), 123
𝒵 𝑝,𝑞 (sheaf of 𝜕-closed (𝑝, 𝑞)-forms), 124
𝒵 𝑝,𝑞 (𝐸) (sheaf of 𝜕-closed 𝐸-valued
(𝑝, 𝑞)-forms), 124
𝑍 𝑝 (𝒰; 𝒮 ) (group of Čech cocycles), 149
Subject Index

Abel, Niels Henrik, 301 compatible with a symplectic form, 287

Abel–Jacobi theorem, 301 Alt convention for wedge products, 229
Abel’s theorem, 311 ample line bundle, 98, 315
abelian variety, 337 analytic continuation of a local isometry, 244
acyclic resolution, 159 analytic function, 12
acyclic sheaf, 159 analytic sheaf, 144
adjoint coherent, 144
formal, 260 analytic variety, 53
of a differential operator, 260 anticanonical bundle, 118
of a finite-dimensional linear map, 258 antiderivation, 103, 113, 264
of a matrix, 68, 259 antiholomorphic tangent bundle, 37
adjoint representation antiholomorphic tangent space, 37
of a Lie algebra, 100 associated line bundle, 91, 94
of a Lie group, 100 associated map, 96
adjunction formula, 120 atlas, 2
affine algebraic variety, 51 holomorphic, 2
nonsingular, 51 smooth, 2
smooth, 51 Aubin, Thierry, 249

of ℂℙ1 , 69
affine coordinates, 6 automorphism, 7

of ℂℙ𝑛 , 310
affine embedding, 55
standard, 55
Akizuki, Yasuo, 319, 321, 322
Akizuki–Nakano identity, 319 Bézout’s theorem, 61
Albanese map, 311 ball, 4
Albanese variety, 298, 300, 301 unit, 4
universal property, 312 base locus, 95
algebraic variety base point, 95
affine, 51 Betti number, 108, 143
nonsingular, 51, 52 Bianchi identity
projective, 52 algebraic, 239
smooth, 51, 52 differential, 219
almost complex manifold, 39 for a connection, 219
almost complex structure, 39 bidegree, 104

351
352 Subject Index

biholomorphic, 8 chain map, 107

biholomorphism, 7 chain rule
local, 7 for holomorphic functions, 31
bilinear relations, 340 for smooth functions, 31
bisectional curvature, holomorphic, 253 chart, 2
Bishop, Errett, 66 chart lemma, 3
blowdown map, 85 Chern, Shiing-Shen, 202
model, 85 Chern class
blowup, 85 of a line bundle, 181
at finitely many points, 87 real, 203
is projective, 335 sheaf-theoretic, 181
Bochner, Salomon, 309, 310 sheaf-theoretic and first real, 204
Bochner vanishing theorem, 309 Chern connection, 208
Borel, Armand, 41 Chern form, 202
boundary, singular, 165 Chow, Wei-Liang, 53
bundle construction theorem, 73 Chow’s theorem, 53, 291

over 𝑀, 25
bundle homomorphism, 25 for hypersurfaces, 291
classification of smooth line bundles, 181
bundle isomorphism, 25 coboundary
holomorphic, 25 Čech, 149
smooth, 25 singular, 166
bundle-valued form, 114, 116 coboundary operator
𝐶 𝑘 manifold, 1
Čech, 148
singular, 166
Calabi, Eugenio, 247
Calabi–Yau manifold, 249, 339 cochain
Calabi–Yau theorem, 247 Čech, 147
canonical bundle, 118 singular, 165
degree of, 253 cochain complex, 107
of projective space, 118 cochain group, 147
canonical curve, 339 cochain homotopy formula, 151
cochain map, 107
on ℂ𝑛 , 4
canonical holomorphic structure
cocompact subgroup, 12
on an open subset of ℂ𝑛 , 4 cocycle
canonical map, 339 Čech, 149
Cartan, Henri, 278 singular, 166
Cartan’s theorems A & B, 278 cocycle condition, 73
category, equivalence of, 133 codimension, 48
Cauchy–Riemann equations, 12, 14 coefficient homomorphism, 166
inhomogeneous, 109 coherent analytic sheaf, 144
Cauchy–Riemann operator, 106 cohomologous, 149
Čech, Eduard, 147 cohomology
Čech coboundary, 149 Čech, 147, 148
Čech cochain, 147 of a cochain complex, 107
Čech cocycle, 149 of constant sheaves, 168
Čech cohomology, 147, 152 sheaf, 147, 152
on an open cover, 148 singular, 166, 168
chain smooth singular, 175
singular, 165 cohomology class, 149

of type (𝑝, 𝑞), 294

smooth, 175 integral, 178, 334
chain complex, 107
chain group, singular, 165 compatible almost complex structure, 287
Subject Index 353

compatible charts conjugate symmetry, 76

holomorphically, 2 conjugation
smoothly, 2 in a complexified vector bundle, 26, 42
compatible with a metric, 195 in a complexified vector space, 24
compatible with the holomorphic structure, in a Lie group, 100
207 of a quaternion, 43
complex-analytic function, 12 of an octonion, 43
complex coordinate frame, 29 connected sum, 100
complex coordinate vector field, 29 connecting homomorphism, 154
complex covector field, 26 in sheaf cohomology, 155
complex curve, 3, 41 connection, 193
complex differential form, 26, 103 Chern, 208
complex dimension, 3 compatible with a metric, 195
complex Grassmannian, 6 compatible with the holomorphic
complex hyperbolic metric, 243, 252 structure, 207
complex Lie group, 10, 51 dual, 218
complex Lie subgroup, 51 metric, 195
complex manifold, 1, 2 tensor product, 219
complex Monge–Ampère equation, 248 connection forms, 194
complex partial derivative, 13 conormal bundle, 119
complex projective space, 5 constant holomorphic sectional curvature,
complex structure 242, 243, 252
on a vector bundle, 35 constant presheaf, 145
on a vector space, 34 constant sheaf, 124
traditional name for holomorphic contractible, 166
structure, 3 locally, 168
complex submanifold, 48 semilocally, 172
embedded, 48 contragredient, 80
immersed, 48 convex, holomorphically, 65
complex surface, 3, 41 convex hull, 65, 69
complex torus, 11, 235 coordinate chart, 2
complex vector bundle, 24 holomorphic, 3
holomorphic, 71 coordinate frame, complex, 29
smooth, 24 coordinate representation, 7
complex vector field, 26 coordinate vector field, complex, 29
complexification coordinates
of a linear map, 23 holomorphic, 3
of a vector bundle, 25 homogeneous, 5
of a vector space, 23 covariant derivative, 193
of the cotangent bundle, 26 along a curve, 197
of the tangent bundle, 26 exterior, 200
complexified cotangent bundle, 26 total, 194
complexified tangent bundle, 26 covering map, 8
component functions of a section, 72 holomorphic, 9
composition smooth, 9
of holomorphic functions, 16 cubic, projective, 57
of holomorphic maps, 8 curvature
of sheaf morphisms, 126 holomorphic bisectional, 253
conformal map, 44 holomorphic sectional, 241
conformal metric, 251 of a connection, 198
conjugate-linear, 24 of a Kähler metric, 240
354 Subject Index

curvature endomorphism field, 239 distribution, 40

curvature forms, 199 divergence, 265
curvature tensor, 239 divisor, 93
curve, complex, 3, 41 effective, 93
cycle, singular, 165 linearly equivalent, 94

𝜕 𝐸 -harmonic, 277
of a meromorphic function, 93

𝜕-Poincaré lemma, 109

of a meromorphic section, 94

𝜕-harmonic form, 276

principal, 94

𝜕-harmonic form, 275

divisor class group, 94

𝜕𝜕-lemma
Dolbeault, Pierre, 105, 173
Dolbeault cohomology, 108
global, 248, 292 functoriality, 108

𝑑-harmonic form, 276

local, 113 with bundle coefficients, 117, 276
Dolbeault complex, 108
de Rham cohomology, 108, 143 Dolbeault Laplacian, 274
with complex coefficients, 143 bundle-valued, 277
de Rham theorem, 176 Dolbeault operator, 105
sheaf-theoretic version, 173 Dolbeault theorem, 173
de Rham–Weil theorem, 159 double skyscraper sheaf, 143, 316
decomposable form, 43 dual bundle, 79
defining function, local, 49 dual connection, 218
degree dual metric, 210, 218, 219
of a codimension-1 variety, 57 dual projective space, 54
of a divisor, 211 dual torus, 311

𝐸-valued form, 114, 116

of a line bundle, 182
of a smooth map, 338
density, Riemannian, 257 effective divisor, 93, 297
determinant bundle, 80 Einstein summation convention, 2, 5, 27, 227
determinant convention for wedge products, elementary symmetric polynomial, 60
229 elliptic complex, 312
difference tensor, 202 elliptic curve, 308
differential Bianchi identity, 219 elliptic differential operator, 268
differential form elliptic integral, 308
bundle-valued, 114, 116 embedded complex submanifold, 48
complex, 103 embedding, 46

of type (𝑝, 𝑞), 104

endomorphism-valued, 114 holomorphic, 46
embedding theorem
real, 103 Kodaira, 330, 334
differential of a smooth map, 29 Stein, 66
differential operator, 259 empty set, sections in a presheaf, 122
elliptic, 268 End(𝐸)-valued form, 114
order zero, 268 endomorphism bundle, 81
dimension, 2 endomorphism-valued form, 114
complex, 3 equivalence of categories, 133
real, 3 étalé space, 129
direct image sheaf, 146 of a presheaf, 129
direct limit, 128 of abelian groups, 129
direct sum of sheaves, 123 of rings, 129
direct system, 128 of vector spaces, 129
directed set, 128 étalé space morphism, 133
discrete Lie group, 9 Euclidean inner product, 4
disk, 4 Euler characteristic of a sheaf, 158, 295
Subject Index 355

fundamental 2-form, 224

Euler’s identity, 63 is homogeneous, 236
evaluation map, 316
evenly covered, 8 fundamental class of a manifold, 165
exact sequence
of abelian groups, 137 GAGA principle, 53
of cochain complexes, 154 Gaussian curvature, 251
of rings, 137 Gaussian integer, 12
of sheaves, 138 Gaussian integers, 41
of vector spaces, 137 genus-degree formula, 254
short, 139 genus of a compact Riemann surface, 183
exceptional hypersurface, 85 geometric normal bundle, 81
exponential sheaf sequence, 142 germ
smooth, 142 of a holomorphic function, 127

global 𝜕𝜕-lemma, 248, 292

extendible section, 197 of a section, 128
exterior covariant differentiation, 200
global differential, 30
face map, 165 global Hodge inner product, 257
Fano manifold, 254 global section, 25
Fermat curve, 59 gluing property of a sheaf, 122
Fermat hypersurface, 59, 250 Godement, Roger, 184
Fermat’s last theorem, 59 Godement resolution, 184, 190
fiber metric, Hermitian, 76 gradient, 265
fiber product, 75, 129, 146 graph of a function, 46, 50
fine sheaf, 163 Grassmannian, 6, 68
is acyclic, 163 Grothendieck, Alexander, 184
finiteness theorem, 278 group action, 9
first real Chern class, 203 free, 9
and sheaf-theoretic Chern class, 204 holomorphic, 9
flabby sheaf, 184 proper, 9
flasque resolution, 184, 190
flasque sheaf, 184 hard Lefschetz theorem, 288
is acyclic, 188 harmonic form, 255, 270, 275
flat connection, 199 Hartogs, Friedrich, 13, 21
flat operator (♭), 229 Hartogs’s extension theorem, 22
form Hermitian adjoint of a matrix, 68, 259
bundle-valued, 114, 116 Hermitian fiber metric, 76
complex, 103 Hermitian inner product, 76

of bidegree (𝑝, 𝑞), 104

endomorphism-valued, 114 Hermitian manifold, 224

of type (𝑝, 𝑞), 104

Hermitian metric, 224
Hermitian vector bundle, 76
formal adjoint, 260 Hilbert space, 258
formally self-adjoint, 270 Hilbert space adjoint, 260
fourfold, 3 Hodge, William V. D., 105, 256, 276, 281
frame, local, 25 Hodge decomposition theorem, 286
Fredholm theorem, 269 Hodge diamond, 287
for elliptic operators, 269 Hodge–Dolbeault theorem, 276
for operators with injective symbol, 312 for bundle-valued forms, 277
free action, 9 Hodge duality, 286
free sheaf, 135 Hodge inner product
locally, 135 global, 257
Fubini, Guido, 235 pointwise, 256
Fubini–Study metric, 235, 334 Hodge Laplacian, 269
356 Subject Index

Hodge metric, 334 holomorphic tangent bundle, 37

of ℂℙ𝑛 , 289
Hodge numbers, 108, 276 holomorphic tangent space, 37
holomorphic vector bundle, 25, 71
of a Riemann surface, 289 holomorphic vector field, 99
Hodge star operator, 262, 263 holomorphically compatible charts, 2
on a Hermitian manifold, 274 holomorphically convex, 65
on a Riemann surface, 274 holonomy group, 254, 339
on a Riemannian manifold, 262 homogeneous coordinates, 5
Hodge theorem, 255 homogeneous function, 83
for elliptic complexes, 313 homogeneous polynomial, 52
for Kähler manifolds, 286 homogenization of a polynomial, 59
for Riemannian manifolds, 272 homologous, 165
Hodge theory, 256 homology
holomorphic action, 9 of a chain complex, 107
holomorphic atlas, 2 singular, 165
holomorphic bisectional curvature, 253 smooth singular, 175
holomorphic bundle isomorphism, 25 homology class, 165
holomorphic coordinate chart, 3 homomorphism bundle, 81
holomorphic coordinates, 3 homomorphism of vector bundles, 25
holomorphic covering map, 9 Hopf manifold, 11, 238
holomorphic embedding, 46 horizontal tangent space, 251
holomorphic form, 118 hull
is closed and harmonic, 292 convex, 65, 69
is parallel, 310 holomorphic, 65
holomorphic function, 7 hyperbolic metric, complex, 243, 252
of one variable, 12 hyperelliptic curve, 338
of several variables, 13 hyperplane, projective, 54
hyperplane at infinity, 55
vector-valued, 13
hyperplane bundle, 90
holomorphic hull, 65
is positive, 320
holomorphic immersion, 46
hypersurface, 48
holomorphic implicit function theorem, 46
in projective space, 291
holomorphic inverse function theorem, 45
holomorphic Jacobian, 31, 37 ideal sheaf, 125, 133, 137
holomorphic manifold, 2 identity theorem, 18
holomorphic map for manifolds, 19

between open subsets of ℂ𝑛 , 2

between manifolds, 7 image of a sheaf morphism, 138
immersed complex submanifold, 48
holomorphic normal bundle, 81 immersion, 46
holomorphic quotient manifold theorem, 10 holomorphic, 46
holomorphic rank theorem, 46 implicit function theorem, 46
holomorphic section, 9, 72 inclusion morphism, 127
holomorphic sectional curvature, 241 induced cohomology homomorphism, 107
constant, 242, 243, 252 injective resolution, 184
holomorphic slice, 48 injective sheaf, 184
holomorphic structure, 2 is acyclic, 188

on ℂ𝑛 , 4
determined by an atlas, 3 is flasque, 185
injective sheaf morphism, 139

on an open subset of ℂ𝑛 , 4
on a manifold, 2 injective symbol, 312
inner product, Hermitian, 76
holomorphic subbundle, 76 integrable almost complex structure, 40
holomorphic submersion, 46 integral cohomology class, 178, 294, 334
Subject Index 357

integration over a smooth chain, 176 Kronecker pairing, 166

interior multiplication, 43, 113, 264 and de Rham cohomology, 178
interior product, see interior multiplication Kummer, Ernst, 250

𝐿2 inner product, 258

inverse function theorem, 45
inverse image sheaf, 146
invertible sheaf, 137 Laplace–Beltrami operator, 265, 308
involutivity, 40 Laplacian
irreducible, 51 Dolbeault, 274
isolated singularities, 22 Hodge, 269
isomorphic bundles, 25 lattice, 11
isomorphic presheaves, 126 Lefschetz, Solomon, 281, 288
isomorphic sheaves, 126 Lefschetz decomposition theorem, 309

Lefschetz theorem on (1, 1)-classes, 294

isomorphism Lefschetz operator, 281
of presheaves, 126
of sheaves, 126 Lefschetz theorem, hard, 288
of vector bundles, 25 Leray, Jean, 121
Iwasawa manifold, 12, 41, 292 Levi-Civita connection, 198, 207
standard, 12 Lie group
complex, 10, 51
Jacobi, Carl Gustav Jacob, 301 discrete, 9
Jacobi inversion theorem, 311 Lie subgroup, complex, 51
Jacobian, holomorphic, 31, 37 lift of a vector field, 61
Jacobian variety, 305, 339 line, projective, 54
line bundle, 25

𝑘-form, complex, 103

K3 surface, 250 associated with a divisor, 94
associated with a hypersurface, 91
Kähler, Erich, 250 classification, 181, 290
Kähler class, 227 degree, 182
Kähler–Einstein metric, 249 negative, 216, 337
Kähler form, 226 on projective space, 290
is harmonic, 309 positive, 216, 320
Kähler identities, 281 linear subspace, projective, 54
for bundle-valued forms, 317 linearly equivalent divisors, 94
Kähler manifold, 226 Liouville’s theorem, 19

on ℂℙ𝑛 , 235 local 𝜕𝜕-lemma, 113

Kähler metric, 226 local biholomorphism, 7

on ℂ𝑛 , 234 local defining function, 49

on a Riemann surface, 235 system of, 91
on complex hyperbolic space, 243, 252 local frame, 25
on complex tori, 235 local isometry, 243
Kähler potential, 234 local operator, 259
Kähler symmetries of the curvature, 240 local section
kernel of a sheaf morphism, 138 of a bundle, 25
Kodaira, Kunihiko, 250, 315, 322 of a continuous map, 123
Kodaira embedding theorem, 330, 334 of a covering map, 9
geometric version, 334 of a map, 47
line bundle version, 330 local trivialization, 24
Kodaira–Nakano–Akizuki vanishing locality property of a sheaf, 122
theorem, 321 locally contractible, 168
Kodaira–Thurston manifold, 288 locally decomposable, 43
Kodaira vanishing theorem, 317, 322 locally finitely generated sheaf, 144
Kronecker homomorphism, 167 locally free sheaf, 135
358 Subject Index

locally zero, 170 order

long exact sequence, 154 of a pole, 93
in sheaf cohomology, 154 of a zero, 13
loop, 254 orientation of a complex manifold, 32
oriented connected sum, 100

𝐶 𝑘, 1 (𝑝, 𝑞)-form, 104

manifold

complex, 1, 2 parallel local frame, 199

smooth, 1, 2 parallel section, 198
topological, 1 along a curve, 197
maximal atlas parallel transport, 197, 254
holomorphic, 2 partition of unity
smooth, 2 sheaf, 163
maximum principle, 19 smooth, 162, 163
meromorphic function, 93 topological, 163
meromorphic section, 93 period lattice, 300
metric connection, 195 period matrix, 300
Möbius transformation, 56 periods of a Riemann surface, 300
model blowdown map, 85 Picard, Charles Émile, 83
Monge–Ampère equation, complex, 248 Picard group, 83, 181
Montel’s theorem, 20 and sheaf cohomology, 179
morphism Picard variety, 182, 298, 301
of presheaves, 126 plane, projective, 54

ℛ-module, 126
of sheaves, 126 Plücker embedding, 69
pluriharmonic function, 119, 189
multiplicity of a zero, 13 Poincaré lemma, 109
musical isomorphism, 229 for the Dolbeault operator, 109
point bundle, 93
Nakano, Shigeo, 319, 321, 322 pointwise Hodge inner product, 256

natural coordinates for 𝑇 ∗ 𝑀, 267

Narasimhan, Raghavan, 66 pole of a meromorphic function, 93

negative (1, 1)-form, 215

polydisk, 4
polynomial, 17

positive (1, 1)-form, 215, 216, 320

negative line bundle, 216, 337 homogeneous, 52
Newlander, August, 40
Newlander–Nirenberg theorem, 40 positive definite, 76

positive (𝑝, 𝑝)-form, 220

Nijenhuis tensor, 42 positive line bundle, 216, 320

nondegenerate 2-form, 227

Nirenberg, Louis, 40
presheaf, 121
nondegenerate bilinear form, 273 constant, 145
nonsingular variety of abelian groups, 122
affine, 51 of rings, 122
projective, 52 of vector spaces, 122
nontrivial section of a vector bundle, 25 that is not a sheaf, 125
norm with respect to a fiber metric, 76 with values in a category, 122
normal bundle presheaf isomorphism, 126
geometric, 81 presheaf morphism, 126
holomorphic, 81 primed summation sign, 105
normal covering map, 8 primitive cohomology class, 309
normal subgroup, 8, 10 principal divisor, 94, 311
principal symbol of a differential operator,
octonions, 43 267
open submanifold, 5, 50 projective algebraic variety, 52
Subject Index 359

nonsingular, 52 Godement, 184, 190

smooth, 52 injective, 184
projective completion, 59 restricting domains or codomains, 50
projective cubic, 57 restriction map in a presheaf, 121
projective hyperplane, 54 restriction of a bundle, 80
projective hypersurface, 291 restriction of a sheaf to an open subset, 123
diffeomorphic, 62 resultant, 60
is connected, 291 of several homogeneous polynomials, 62
projective line, 54 Ricci curvature, 245
projective linear subspace, 54 Ricci flat, 250
projective manifold, 52 Ricci form, 247
projective plane, 54 and first Chern class, 247
projective quadric, 57 Riemann, Bernhard, 297
projective quartic, 57 Riemann bilinear relations, 340
projective quintic, 57 Riemann curvature tensor, 239
projective space, complex, 5 Riemann form, 336
projective tangent space, 55 Riemann–Roch theorem, 295, 297
projective transformation, 47 Riemann sphere, 56
projectively equivalent, 54 Riemann surface, 41
projectivization of a vector space, 6 Riemannian covering, 244
proper action, 9 Riemannian density, 257
pseudoconvex, 67 Riemannian submersion, 251
pullback bundle, 75, 210 ring
pullback metric, 210 has multiplicative identity, 125
pullback of sections, 75, 76 presheaf of, 122
sheaf of, 125
quadric, projective, 57 Roch, Gustav, 297
quartic, projective, 57 rough section, 25, 129
quaternions, 43 of a continuous map, 123
quintic, projective, 57
quotient bundle, 80 saturated subset, 63
quotient manifold theorem, 10 scalar curvature, 245
quotient sheaf, 133 section

ℛ-module morphism, 126

holomorphic, 72
local, 25
rank of a quadratic polynomial, 57 of a bundle, 25
rank theorem, 46 of a continuous map, 123
rank-nullity law, 158 of a presheaf, 121
rational function is holomorphic, 17 of a vector bundle along a curve, 197
real-analytic manifold, 1 over the empty set, 122
real Chern class, 203 rough, 25, 129
real differential form, 103 sectional curvature, holomorphic, 241, 242
real dimension, 3 Segre embedding, 68
real vector, 24 self-adjoint, formally, 270
refinement of an open cover, 150 semilocally contractible, 172
refining map, 150 separates directions, 64, 97
regular level set, 50 separates points, 64, 97
regular point, 51 Serre, Jean-Pierre, 41, 53, 278
regular value, 50 Serre duality, 278
resolution of a sheaf, 159 sesquilinearity, 76
acyclic, 159 sharp operator (♯), 229
flasque, 184, 190 sheaf, 122
360 Subject Index

flabby, 184 singular simplex, 164

flasque, 184 smooth, 175
free, 135 singularity, isolated, 22
injective, 184 skyscraper sheaf, 124
invertible, 137 double, 143, 316
locally free, 135 slice, 48

of ℰ -modules, 125
of discontinuous sections, 187 slice chart, 48
smooth atlas, 2

of 𝒪-modules, 125
of modules over a sheaf of rings, 125 smooth bundle isomorphism, 25

of ℛ-modules, 125
smooth chain, 175
smooth complex vector bundle, 24
of rings, 125 smooth covering map, 9

of ℤ-modules, 126
of rough sections, 186 smooth exponential sheaf sequence, 142
smooth manifold, 1, 2
sheaf cohomology, 152 smooth map, 2
Čech, 147 smooth partition of unity, 162, 163
functoriality, 153 smooth singular chain, 175
long exact sequence, 154 smooth singular cohomology, 175
sheaf isomorphism, 126 smooth singular homology group, 175
sheaf morphism, 126 smooth singular simplex, 175
composition, 126 smooth structure, 2
injective, 139 smooth variety
surjective, 139 affine, 51
sheaf partition of unity, 163 projective, 52
sheaf-theoretic Chern class, 181 smoothly compatible charts, 2
and first real Chern class, 204 soft sheaf, 190
sheafification, 131 stalk
universal property, 131 of a presheaf, 128
short exact sheaf sequence, 139 of an étalé space, 129
sign of a permutation, 257 of the sheaf of holomorphic functions, 127
simple zero, 13 stalk homomorphism, 128

standard basis of ℂ𝑛 , 35
simplex standard affine embedding, 55
singular, 164
smooth singular, 175 standard holomorphic structure on a vector
standard, 164 space, 5

standard metric on ℂ𝑛 , 225, 234

simplicial cohomology, 191 standard Iwasawa manifold, 12
simplicial complex, 191
singular boundary operator, 165 standard simplex, 164
singular chain, 165 standard symplectic form, 225
smooth, 175 star of a simplex, 191
singular chain group, 165 star operator, see Hodge star operator
singular coboundary, 166 Stein embedding theorem, 66
singular coboundary operator, 166 Stein manifold, 65
singular cochain, 165 Study, Eduard, 235
singular cochain group, 165 subbundle, 76
singular cocycle, 166 submanifold, 48
singular cohomology, 166 complex, 48
singular cycle, 165 embedded, 48
singular homology, 165 immersed, 48
smooth, 175 open, 5
singular point of a variety, 51 submersion, 46
Subject Index 361

holomorphic, 46 unit disk, 4

subpresheaf, 123 unitary group, 68
subsheaf, 123 universal coefficient theorem, 167
summation convention, 2, 5, 27, 227 universal property
superstring theory, 250 of sheafification, 131
support of a sheaf morphism, 163 of the Albanese variety, 312
surface
complex, 3, 41 vanishing
Riemann, 41 simply, 90
surjective sheaf morphism, 139 to second order, 124, 137, 316
symbol of a differential operator variety
principal, 267 affine, 51
total, 266 algebraic, 51, 52
symmetric polynomial, 60 analytic, 53
fundamental theorem on, 60 determined by a section, 90
symplectic form, 227, 287 nonsingular, 51, 52
standard, 225 projective, 52
symplectic manifold, 287 smooth, 51, 52
system of local defining functions, 91 vector bundle
complex, 24
tangent bundle Hermitian, 76
antiholomorphic, 37 holomorphic, 25, 71
holomorphic, 37 smooth, 24
tautological bundle, 84, 100 vector field, holomorphic, 99
Taylor series, 18 vector space, holomorphic structure on, 5
tensor product bundle, 80 very ample line bundle, 97, 315
tensor product connection, 219
wedge product
tensor product metric, 210
Alt convention, 229
tensor product sheaf, 134
determinant convention, 229
threefold, 3
with a bundle-valued form, 114, 115
Thurston, William, 287, 288
with an endomorphism-valued form, 115
topological manifold, 1
Weil, André, 250
topological partition of unity, 163
Whitney embedding theorem, 64
torsion-free connection, 207 Whitney sum, 80
torsion subgroup, 179
torus, complex, 11, 235 Yau, Shing-Tung, 247, 249
total covariant derivative, 194
total derivative, 29 Zariski topology, 53, 183
total symbol of a differential operator, 266 zero section, 25
transition function, 2, 71 zigzag lemma, 154
transverse to a submanifold, 50
triangulation, 165
smooth, 165
trivial sheaf, 139
trivial vector bundle, 25
trivializing cover, 25
type of a differential form, 104

𝒰-small chain, 171

𝒰-small simplex, 171
uniformization theorem, 249
unit ball, 4
 SELECTED PUBLISHED TITLES IN THIS SERIES

244 John M. Lee, Introduction to Complex Manifolds, 2024

238 Julio González-Dı́az, Ignacio Garcı́a-Jurado, and M. Gloria Fiestras-Janeiro,
An Introductory Course on Mathematical Game Theory and Applications, Second Edition,
2023
237 Michael Levitin, Dan Mangoubi, and Iosif Polterovich, Topics in Spectral
Geometry, 2023
235 Bennett Chow, Ricci Solitons in Low Dimensions, 2023
234 Andrea Ferretti, Homological Methods in Commutative Algebra, 2023
233 Andrea Ferretti, Commutative Algebra, 2023
232 Harry Dym, Linear Algebra in Action, Third Edition, 2023
231 Luı́s Barreira and Yakov Pesin, Introduction to Smooth Ergodic Theory, Second
Edition, 2023
230 Barbara Kaltenbacher and William Rundell, Inverse Problems for Fractional Partial
Differential Equations, 2023
229 Giovanni Leoni, A First Course in Fractional Sobolev Spaces, 2023
228 Henk Bruin, Topological and Ergodic Theory of Symbolic Dynamics, 2022
227 William M. Goldman, Geometric Structures on Manifolds, 2022
226 Milivoje Lukić, A First Course in Spectral Theory, 2022
225 Jacob Bedrossian and Vlad Vicol, The Mathematical Analysis of the Incompressible
Euler and Navier-Stokes Equations, 2022
224 Ben Krause, Discrete Analogues in Harmonic Analysis, 2022
223 Volodymyr Nekrashevych, Groups and Topological Dynamics, 2022
222 Michael Artin, Algebraic Geometry, 2022
221 David Damanik and Jake Fillman, One-Dimensional Ergodic Schrödinger Operators,
2022
220 Isaac Goldbring, Ultrafilters Throughout Mathematics, 2022
219 Michael Joswig, Essentials of Tropical Combinatorics, 2021
218 Riccardo Benedetti, Lectures on Differential Topology, 2021
217 Marius Crainic, Rui Loja Fernandes, and Ioan Mărcuţ, Lectures on Poisson
Geometry, 2021
216 Brian Osserman, A Concise Introduction to Algebraic Varieties, 2021
215 Tai-Ping Liu, Shock Waves, 2021
214 Ioannis Karatzas and Constantinos Kardaras, Portfolio Theory and Arbitrage, 2021
213 Hung Vinh Tran, Hamilton–Jacobi Equations, 2021
212 Marcelo Viana and José M. Espinar, Differential Equations, 2021
211 Mateusz Michalek and Bernd Sturmfels, Invitation to Nonlinear Algebra, 2021
210 Bruce E. Sagan, Combinatorics: The Art of Counting, 2020
209 Jessica S. Purcell, Hyperbolic Knot Theory, 2020
208 Vicente Muñoz, Ángel González-Prieto, and Juan Ángel Rojo, Geometry and
Topology of Manifolds, 2020
207 Dmitry N. Kozlov, Organized Collapse: An Introduction to Discrete Morse Theory, 2020
206 Ben Andrews, Bennett Chow, Christine Guenther, and Mat Langford, Extrinsic
Geometric Flows, 2020
205 Mikhail Shubin, Invitation to Partial Differential Equations, 2020
204 Sarah J. Witherspoon, Hochschild Cohomology for Algebras, 2019
203 Dimitris Koukoulopoulos, The Distribution of Prime Numbers, 2019

For a complete list of titles in this series, visit the

AMS Bookstore at [Link]/bookstore/gsmseries/.
Complex manifolds are smooth manifolds endowed with coordinate charts that
overlap holomorphically. They have deep and beautiful applications in many
areas of mathematics. This book is an introduction to the concepts, techniques,
and main results about complex manifolds (mainly compact ones), and it tells a
story. Starting from familiarity with smooth manifolds and Riemannian geometry,
it gradually explains what is different about complex manifolds and develops
most of the main tools for working with them, using the Kodaira embedding
theorem as a motivating project throughout.
The approach and style will be familiar to readers of the author’s previous graduate
texts: new concepts are introduced gently, with as much intuition and motiva-
tion as possible, always relating new concepts to familiar old ones, with plenty
of examples. The main prerequisite is familiarity with the basic results on topo-
logical, smooth, and Riemannian manifolds. The book is intended for graduate
students and researchers in differential geometry, but it will also be appreciated
by students of algebraic geometry who wish to understand the motivations, anal-
ogies, and analytic results that come from the world of differential geometry.

For additional information

and updates on this book, visit
[Link]/bookpages/gsm-244

GSM/244
[Link]